Website Index

• Aaron Kiely
• k very highly of them: Mike Burl, Aaron Kiely, Matt Klimesh and Sam Dolinar. Ho
• Absolute
• d encode three type of constraints: Absolute Position Relative Position Relative
• sition of the robot is given by the Absolute Position constraint, commonly assig
• Agency
• he Defense Advanced Research Project Agency - DARPA - was interested in understa
• he Defense Advanced Research Project Agency - DARPA - was interested in understa
• Albert Einstein
• and white list the combination "Albert Einstein". In this way both "Albert Eins
• rt Einstein". In this way both "Albert Einstein" and "Einstein" would be entrie
• Anatole Litvak
• eded in the text of the message. Anatole Litvak directed two movies in 1937: "To
• Andres Castano
• Andres Castano Last Position NASA / Jet Propuls
• Andres Castano / others The DARPA Shredder Chal
• Andres Castano / others / Udacity Autonomous Ca
• Andres Castano / others / Udacity Search Engine
• Andres Castano / others / DARPA Shredder Challe
• Apple
• came and pasted them in Keynote, the Apple version of Power Point. In hindsigh,
• Attribution-NoDerivs-NonCommercial
• der a Creative Commons Attribution-NoDerivs-NonCommercial 1.0 Generic License .
• Attribution-NonCommercial-ShareAlike
• er a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 United States Lic
• Autonomous
• osed the class. Udacity - CS-373 - Autonomous car - post on SLAM I took Udacity
• Andres Castano / others / Udacity Autonomous Car Quick calculation of the and m



• Background
• check if the matrices are correct. Background I took Udacity's CS-373 class - Pr
• Becky
• f while my children grow up. My wife Becky and I have three children: Sofi, now
• Bella
• Sofi, now 4 1/2, and twins Danny and Bella, age 3, all adopted. Cosita, our Shi
• h all of them but in particular with Bella who, injured with an in-utero stroke
• Block
• nge everytime that a change is made. Block diagram of Index Maker Index Maker i



• Challenge
• astano / others The DARPA Shredder Challenge The Defense Advanced Research Proj
• d 1st and 2nd in the DARPA's Grand Challenge and Urban Challenge , respectively
• DARPA's Grand Challenge and Urban Challenge , respectively. The students of th
• d 1st and 2nd in the DARPA's Grand Challenge and Urban Challenge, respectively.
• DARPA's Grand Challenge and Urban Challenge, respectively. The student of this
• Castano / others / DARPA Shredder Challenge How do you measure solving the "ea
• x puzzles. They issue the Shredder Challenge , a contest to put together, from
• Cienfuegos
• ocation was the Jose Marti park in Cienfuegos. A specific search for the park sh
• owed that the dome was that of the Cienfuegos City Hall. After the contest I fou
• City
• t the dome was that of the Cienfuegos City Hall. After the contest I found out t
• Clearly
• not it. I continued using Keynote. Clearly it was not the best tool for this t
• Comment
• sum of the weighted outgoing edges: Comment on the solution This example was gi
• Commons
• s work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlik
• s work is licensed under a Creative Commons Attribution-NoDerivs-NonCommercial
• Comrade
• ved". The greeting in the message, "Comrade", was the clue that "Tovarich" was
• Continue
• ou what to do with the first three. Continue the pattern with the last three. Th
• Cosita
• Danny and Bella, age 3, all adopted. Cosita, our Shih-Tzu, rounds up the family.
• Crawl_web_by_keyword
• purple in the block diagram: Crawl_web_by_keyword Suggest_index_entries select_
• Create_html_index
• x_entries select_index_entries Create_html_index The input to these functions a
• Creative
• ario. This work is licensed under a Creative Commons Attribution-NonCommercial-S
• file. This work is licensed under a Creative Commons Attribution-NoDerivs-NonCom
• Cuba
• ? The country is easy enough to spot: Cuba. However, the city/town is very diffi
• to clue you that the map was that of Cuba. The city has to be deduced exclusive
• tatue. An image search for statues in Cuba gave me one in which you could see, o



• DARPA
• Andres Castano / others The DARPA Shredder Challenge The Defense Advan
• e Advanced Research Project Agency - DARPA - was interested in understanding ho
• omous cars placed 1st and 2nd in the DARPA's Grand Challenge and Urban Challeng
• omous cars placed 1st and 2nd in the DARPA's Grand Challenge and Urban Challeng
• Andres Castano / others / DARPA Shredder Challenge How do you measur
• e Advanced Research Project Agency - DARPA - was interested in understanding ho
• Danny
• children: Sofi, now 4 1/2, and twins Danny and Bella, age 3, all adopted. Cosit
• Dave
• w of these distinctive chads and got 'Dave' and 'George' in a matter of a few ho
• rators? I got 3 pts for 'George' and 'Dave'. I would have gotten 2 more for 'PK'
• Deal
• functions that will create the index. Deal with the retrieval of pages for which
• Defense
• rs The DARPA Shredder Challenge The Defense Advanced Research Project Agency -
• ng the questions in red Summary The Defense Advanced Research Project Agency -
• Distances
• ute position constraint. The setup Distances are shown in blue, weights are sho
• Dolinar
• , Aaron Kiely, Matt Klimesh and Sam Dolinar. However, they entered the contest
• Download
• ngine functions seen in the course. Download code Examples: index of this websit



• Einstein
• example, if the words "Albert" and "Einstein" are in the suggested word list, it
• white list the combination "Albert Einstein". In this way both "Albert Einstein
• Einstein". In this way both "Albert Einstein" and "Einstein" would be entries bu
• his way both "Albert Einstein" and "Einstein" would be entries but "Albert", by
• Engine
• e contest. Udacity - CS-101 - Search Engine - Index Maker I took Udacity 's CS-1
• 's CS-101 class - Building a Search Engine - in March 2012. Initially I took th
• es Castano / others / Udacity Search Engine Index Maker - a script to create ind
• Examples
• s seen in the course. Download code Examples: index of this website index of CS-



• Family
• our Shih-Tzu, rounds up the family. Family matters We wanted our kids to learn
• Finally
• n of the robot and the measurement. Finally, depending on our confidence on a p
• bservation from the first location. Finally, the robot moves +3 from which it o
• use uppercase tags and attributes. Finally, searching for "href" itself leads



• Generic
• ribution-NoDerivs-NonCommercial 1.0 Generic License .
• George
• istinctive chads and got 'Dave' and 'George' in a matter of a few hours. I also
• the collaborators? I got 3 pts for 'George' and 'Dave'. I would have gotten 2 m
• Goldsong
• zle. I participated using the name 'Goldsong'. This page describes my experience
• nts. I participated using the name 'Goldsong' and managed to place 10th. This is
• Google
• to approach a large corporation like Google or Yahoo to create a simple index of
• Grand
• rs placed 1st and 2nd in the DARPA's Grand Challenge and Urban Challenge , resp
• rs placed 1st and 2nd in the DARPA's Grand Challenge and Urban Challenge, respe
• Graph-SLAM
• alculation of the and matrices for Graph-SLAM The following is a quick way to de
• s it explores its surroundings. In Graph-SLAM, the version of SLAM taught in CS-
• Grove
• obility and Robotic Systems 4800 Oak Grove Dr. Pasadena, CA 91109 my JPL web pa



• HTML
• create indexes from online documents: HTML indexer , a commercial package that r
• the four routines, shown in red, is a HTML file ready to be uploaded to the webs
• t online. By default, it is a strict XHTML 1.0 page in the english language that
• Hall
• dome was that of the Cienfuegos City Hall. After the contest I found out that t
• Hash
• resting and I even got to learn about Hash tables, a data structure that I had n
• Hopefully
• hat many people found interesting. Hopefully, somebody else will find it useful
• is enough to write such interface. Hopefully, I will be able to after taking th



• Ideally
• eady to be uploaded to the website. Ideally, the user would have had access to
• Index
• . Udacity - CS-101 - Search Engine - Index Maker I took Udacity 's CS-101 class
• ano / others / Udacity Search Engine Index Maker - a script to create indexes o
• forum entry for the contest What is Index Maker Index Maker is a Python script
• for the contest What is Index Maker Index Maker is a Python script that semi-a
• t a change is made. Block diagram of Index Maker Index Maker is composed of fun
• s made. Block diagram of Index Maker Index Maker is composed of functions, show
• Initial
• constrains shown on the graph are: Initial position: x 0 = -3 Relative measure
• Intranets
• particular word or topic is found. Intranets and business websites offen attemp



• JGOMEZ
• Z" Hence, the deciphered message is "JGOMEZ" Puzzle 3 2 pts: What is the indicat
• Java
• re with versions for both Windows and Java. I think that a new program can easil
• Jet Propulsion Laboratory
• stano Last Position NASA / Jet Propulsion Laboratory Senior Member of Technical
• Jose Marti
• a band stand. The location was the Jose Marti park in Cienfuegos. A specific sea



• Keynote
• which they came and pasted them in Keynote, the Apple version of Power Point.
• gh, my selection of tools was poor. Keynote is a very good presentation tool bu
• this was not it. I continued using Keynote. Clearly it was not the best tool f
• spend less time pulling pieces into Keynote and more time pondering how to conn



• La Havana
• that they were the coordinates of La Havana and Nassau, meant to clue you that
• License
• ercial-ShareAlike 3.0 United States License .
• -NoDerivs-NonCommercial 1.0 Generic License .
• Localization
• ful. SLAM stands for Simultaneous Localization and Mapping. It is a technique th



• Maker
• ity - CS-101 - Search Engine - Index Maker I took Udacity 's CS-101 class - Bui
• others / Udacity Search Engine Index Maker - a script to create indexes of intr
• entry for the contest What is Index Maker Index Maker is a Python script that
• he contest What is Index Maker Index Maker is a Python script that semi-automat
• ange is made. Block diagram of Index Maker Index Maker is composed of functions
• . Block diagram of Index Maker Index Maker is composed of functions, shown in p
• March
• lass - Building a Search Engine - in March 2012. Initially I took this class be
• ass - Programming a Robotic Car - in March 2012. The class was taught by Sebast
• ass - Programming a Robotic Car - in March 2012. The class was taught by Sebast
• Matt Klimesh
• of them: Mike Burl, Aaron Kiely, Matt Klimesh and Sam Dolinar. However, they en
• Measurement
• sition Relative Position Relative Measurement of a landmark The only known posi
• Member
• A / Jet Propulsion Laboratory Senior Member of Technical Staff Mobility and Robo
• Memoization
• to the copyright or the glosary. Memoization of all the visited pages to speed
• Mike Burl
• ers and think very highly of them: Mike Burl, Aaron Kiely, Matt Klimesh and Sam
• Mobility
• ry Senior Member of Technical Staff Mobility and Robotic Systems 4800 Oak Grove
• Mosaic
• above, in NCSA, they were developing Mosaic, the first browser and the base code



• NASA
• Andres Castano Last Position NASA / Jet Propulsion Laboratory Senior Me
• NCSA
• ute in Urbana, three floors above, in NCSA, they were developing Mosaic, the fir
• Nassau
• ere the coordinates of La Havana and Nassau, meant to clue you that the map was
• Netscape
• browser and the base code for both Netscape IE. I had a web page up and running



• Omega
• node i and node j: Thus we have the Omega matrix: Also, an easy way to verify



• Paper-bound
• u do not know what to search for. Paper-bound books give a broad-topic overview
• Pasadena
• Robotic Systems 4800 Oak Grove Dr. Pasadena, CA 91109 my JPL web page You can c
• Personally
• in how you figured out the answer. Personally, I found the answering of many of
• Points
• ch one progressively more difficult. Points were awarded, not for solving the pu
• Position
• Andres Castano Last Position NASA / Jet Propulsion Laboratory Se
• three type of constraints: Absolute Position Relative Position Relative Measurem
• traints: Absolute Position Relative Position Relative Measurement of a landmark
• the robot is given by the Absolute Position constraint, commonly assigned to th
• Power
• hem in Keynote, the Apple version of Power Point. In hindsigh, my selection of
• Presumably
• /or initials of the collaborators? Presumably, the "collaborators" were the peop
• Project
• lenge The Defense Advanced Research Project Agency - DARPA - was interested in
• mmary The Defense Advanced Research Project Agency - DARPA - was interested in
• Puzzle
• nce with the puzzles and my results. Puzzle 1 Like many other contestants, I fou
• s my results at the end of the page. Puzzle 2 The second puzzle had about 350 ch
• wering the question gave you points. Puzzle 3 Puzzles 1 and 2 definitely did no
• e question gave you points. Puzzle 3 Puzzles 1 and 2 definitely did no need a co
• a computational tool to solve them. Puzzles 4 and 5 definitely were going to ne
• 5 definitely were going to need it. Puzzle 3 was an unknown. Should I solve it
• was a pattern that appeared useful. Puzzle 1 had 150 useful chads and it took a
• solve (or at least it should have). Puzzle 2 had 300 useful chads and took abou
• useful chads and took about 20 hrs. Puzzle 3 had about 600 useful chads. It app
• from the partially-assembled puzzle? Puzzle 4 It did not look sensible to attack
• 50 pts. earned by the winning team. Puzzle 1 2 pts: What is the appropriate tit
• t "Tovarich" was the correct answer. Puzzle 2 4 pts: What is the deciphered mess
• , the deciphered message is "JGOMEZ" Puzzle 3 2 pts: What is the indicated count
• he image of the park from Wikipedia. Puzzle 4 12 pts: What are the name and/or i
• Puzzles
• question gave you points. Puzzle 3 Puzzles 1 and 2 definitely did no need a co
• a computational tool to solve them. Puzzles 4 and 5 definitely were going to ne
• Python
• ed to take CS-373 but I did not know Python (which CS-101 was going to teach) an
• What is Index Maker Index Maker is a Python script that semi-automates the creat
• e course and neither my knowledge of Python nor html is enough to write such int



• Rappers
• ouch with another team, the "Herded Rappers", made of people from JPL. I know f
• likely that we would have won. The "Rappers", like the winning and many of the
• Relative
• e of constraints: Absolute Position Relative Position Relative Measurement of a
• Absolute Position Relative Position Relative Measurement of a landmark The only
• aph are: Initial position: x 0 = -3 Relative measurement: L = x 0 +10 Relative p
• 3 Relative measurement: L = x 0 +10 Relative pose: x 1 = x 0 +5 Relative measure
• x 0 +10 Relative pose: x 1 = x 0 +5 Relative measurement: L = x 1 +5 Relative po
• +5 Relative measurement: L = x 1 +5 Relative pose: x 2 = x 1 +3 Relative measure
• x 1 +5 Relative pose: x 2 = x 1 +3 Relative measurement: L = x 2 +1. Weight = 5
• Research
• dder Challenge The Defense Advanced Research Project Agency - DARPA - was intere
• in red Summary The Defense Advanced Research Project Agency - DARPA - was intere
• Restrict
• make to the crawl_web routine are: Restrict the pages visited using white- and
• Results
• een able to solve puzzle 5 on time. Results I got a total of 17 pts. I could ha
• Robotic
• ber of Technical Staff Mobility and Robotic Systems 4800 Oak Grove Dr. Pasadena
• ity 's CS-373 class - Programming a Robotic Car - in March 2012. The class was
• city's CS-373 class - Programming a Robotic Car - in March 2012. The class was



• SLAM
• y - CS-373 - Autonomous car - post on SLAM I took Udacity 's CS-373 class - Prog
• reproducing a post in the forum about SLAM that many people found interesting. H
• ulation of the and matrices for Graph-SLAM The following is a quick way to deriv
• forum. Here I reproduce a post about SLAM that many people found useful, in the
• Udacity students also find it useful. SLAM stands for Simultaneous Localization
• t explores its surroundings. In Graph-SLAM, the version of SLAM taught in CS-373
• ndings. In Graph-SLAM, the version of SLAM taught in CS-373, the robot motions a
• Sam Dolinar
• rl, Aaron Kiely, Matt Klimesh and Sam Dolinar. However, they entered the contes
• Science
• intervention is no longer possible. Science Engineering I work in robotics and
• Search
• with the contest. Udacity - CS-101 - Search Engine - Index Maker I took Udacity
• Udacity 's CS-101 class - Building a Search Engine - in March 2012. Initially I
• Andres Castano / others / Udacity Search Engine Index Maker - a script to cre
• Sebastian Thrun
• h 2012. The class was taught by Sebastian Thrun, one of the most successful rob
• h 2012. The class was taught by Sebastian Thrun, one of the most successful rob
• Senior
• ion NASA / Jet Propulsion Laboratory Senior Member of Technical Staff Mobility a
• Shih-Tzu
• la, age 3, all adopted. Cosita, our Shih-Tzu, rounds up the family. Family matte
• Shredder
• Andres Castano / others The DARPA Shredder Challenge The Defense Advanced Rese
• Andres Castano / others / DARPA Shredder Challenge How do you measure solvin
• lve complex puzzles. They issue the Shredder Challenge , a contest to put togeth
• Simultaneous
• o find it useful. SLAM stands for Simultaneous Localization and Mapping. It is a
• Social
• rocessing, image understanding, etc. Social issues I am an acid critic of our so
• Sofi
• wife Becky and I have three children: Sofi, now 4 1/2, and twins Danny and Bella
• Staff
• aboratory Senior Member of Technical Staff Mobility and Robotic Systems 4800 Oa
• States
• -NonCommercial-ShareAlike 3.0 United States License .
• Suggest_index_entries
• iagram: Crawl_web_by_keyword Suggest_index_entries select_index_entries Create_
• Summary
• out answering the questions in red Summary The Defense Advanced Research Proje
• Systems
• echnical Staff Mobility and Robotic Systems 4800 Oak Grove Dr. Pasadena, CA 911



• Technical
• ulsion Laboratory Senior Member of Technical Staff Mobility and Robotic Systems
• Tovarich
• itvak directed two movies in 1937: "Tovarich" and "The woman I loved". The greet
• sage, "Comrade", was the clue that "Tovarich" was the correct answer. Puzzle 2 4



• Udacious
• tine is the crawl_web function from Udacious. We saw three versions of it: indis
• Udacity
• bes my experience with the contest. Udacity - CS-101 - Search Engine - Index Ma
• Search Engine - Index Maker I took Udacity 's CS-101 class - Building a Search
• open contest that closed the class. Udacity - CS-373 - Autonomous car - post on
• utonomous car - post on SLAM I took Udacity 's CS-373 class - Programming a Rob
• Andres Castano / others / Udacity Autonomous Car Quick calculation of
• ices are correct. Background I took Udacity's CS-373 class - Programming a Robo
• useful, in the hope that new CS-373 Udacity students also find it useful. SLAM
• Andres Castano / others / Udacity Search Engine Index Maker - a scrip
• Unfortunatelly
• all within a level of the crawl. Unfortunatelly, none of this versions is useful
• e, I was 5th in the leaderboard. Unfortunatelly, life caught up with me and the
• Urban
• d in the DARPA's Grand Challenge and Urban Challenge , respectively. The studen
• d in the DARPA's Grand Challenge and Urban Challenge, respectively. The student
• Urbana
• st floor of the Beckman Institute in Urbana, three floors above, in NCSA, they w



• Weight
• +3 Relative measurement: L = x 2 +1. Weight = 5 In this example taken straight f
• Wikipedia
• ased on the image of the park from Wikipedia. Puzzle 4 12 pts: What are the nam
• Windows
• , a commercial package that runs on Windows, and XRefHT , a freeware with versi
• , a freeware with versions for both Windows and Java. I think that a new progra



• XHTML
• t online. By default, it is a strict XHTML 1.0 page in the english language tha
• XRefHT
• al package that runs on Windows, and XRefHT , a freeware with versions for both



• Yahoo
• h a large corporation like Google or Yahoo to create a simple index of their we



• absolute
• y an amount d 0 1 . Since x 0 is an absolute reference, this would mean that the
• an that the location of x 1 is also absolute. The problem is that there is an er
• olve for the vector X to obtain the absolute locations of the robot, i.e., the a
• espect to the location set with the absolute position constraint. The setup Dist
• access
• ely. The students of this course had access to a very active forum that probably
• vely. The student of this course had access to a very active forum. Here I repro
• te. Ideally, the user would have had access to the parameter files via a GUI or,
• acid
• rstanding, etc. Social issues I am an acid critic of our society. We have it all
• actions
• n now, when the consequences of our actions no longer need to be predicted, we
• active
• of this course had access to a very active forum that probably will become priv
• of this course had access to a very active forum. Here I reproduce a post about
• activities
• long-term common ones. Others Some activities carried out during the copious fre
• actual
• te locations of the robot, i.e., the actual path that it followed, and the locat
• adapt
• lusively a given domain so we had to adapt a little. The main changes that we h
• addition
• ar old and they learn it better. In addition, it is fun, gets you closer to your
• additional
• s. if I had realized that I had an additional answer in puzzle 4. Thus, the high
• adjecent
• ly configurations of chads that are adjecent to a particular one. In this case,
• allow
• user does not care for. The function allows the user to alter the suggested ent
• tomatically, after which the user is allow to interact with the list and add or
• allows
• s, prepares them for the future, and allows you to identify learning dissabiliti
• and Mapping. It is a technique that allows a robot to use repeated observations
• user does not care for. The function allows the user to alter the suggested entr
• alone
• . Also, in spite that I like to work alone, it was obvious that it would have b
• alter
• for. The function allows the user to alter the suggested entries in two ways. F
• modifying the html page or, without altering the page by modifying a CSS file.
• alternatively
• and measurement constraints or, alternatively, to use the graph to check if th
• amount
• robot moves to a new location by an amount d 0 1 . Since x 0 is an absolute ref
• unusual for a user to spend a large amount of time traversing the tree, trying
• anchor
• ef=" in insuficient. For example, an anchor tag does not need to have the "href"
• answer
• " portion of the challenge? Find out answering the questions in red Summary The
• be completed to the point needed to answer an associated question and to explai
• d to explain how you figured out the answer. Personally, I found the answering o
• the answer. Personally, I found the answering of many of the questions more tri
• he image). What would have been your answer to the question? I'll discuss my res
• the puzzle gave you no points. Only answering the question gave you points. Puz
• solve, pretty doable. Thus, I had my answer and Boy, was I wrong! At the beginni
• uzzle. However, I had no idea how to answer the questions associated with the pu
• 00 hrs. to assemble enough pieces to answer the two questions of this puzzle: Wh
• tants, these few chads contained the answers to the question associated with the
• ize it, so I did not submit it as an answer and did not get credit for it. Can y
• wise, I did not realize that all the answers stood out from the general content
• tool, it was completely feasible to answer the questions without assembling it
• ad realized that I had an additional answer in puzzle 4. Thus, the highest that
• clue that "Tovarich" was the correct answer. Puzzle 2 4 pts: What is the deciphe
• answers
• ants, these few chads contained the answers to the question associated with the
• ise, I did not realize that all the answers stood out from the general content
• appereance
• dation test with a good grade. The appereance of the page can be changed by hand
• application
• ndexing. The target user for this application would be one that does not want t
• approach
• would be one that does not want to approach a large corporation like Google or
• nnect them. There were many ways to approach this puzzle: assemble the coffee mu
• beautiful part is that all of these approaches worked equally well. In the end,
• approaches
• eautiful part is that all of these approaches worked equally well. In the end, i
• appropriate
• zle, the question was What is the appropriate title being referenced? Well.. I
• team. Puzzle 1 2 pts: What is the appropriate title being referenced? You neede
• approximate
• but I was expecting it to find an approximate solution that could be useful. Th
• approximation
• that the relationship is only an approximation because of the errors in both th
• arbitray
• ne will return a valid result to an arbitray query, i.e., "a", "as", "at", this
• area
• gram can easily make an impact in the area of indexing. The target user for this
• artificial
• to intelligence, both natural and artificial, e.g., brain organization, theorie
• assemble
• m and thus, using it as a canvas to assemble the chads was slow. It took me abou
• as slow. It took me about 20 hrs to assemble enough chads to be able to read the
• ther a computatinal tool to help me assemble the puzzle but this was not it. I c
• many ways to approach this puzzle: assemble the coffee mug stain, assemble the
• zle: assemble the coffee mug stain, assemble the borders, assemble the chads wit
• ee mug stain, assemble the borders, assemble the chads with red ink, and more. T
• he right choice. After 40 hrs I had assembled about 80% of the puzzle. However,
• he end it took me about 100 hrs. to assemble enough pieces to answer the two que
• figure them out from the partially-assembled puzzle? Puzzle 4 It did not look s
• the puzzle. Hence, I did not try to assemble the words with distinctive hand wri
• ke! Although it was not possible to assemble the puzzle without the tool, it was
• referred to a church. Eventually, I assembled enough chads to recognize the stru
• assign
• solute Position constraint, commonly assigned to the initial position of the rob
• x 0 , although I believe it could be assigned to any one position. Suppose that
• = x i + k) or W*L j = W*x i + W*k We assign a large weigh to an equation (i.e.,
• arge error (high variance) should be assigned a low weight (0 W 1). The vector e
• ors. This result is a consequence of assigning a large weight to the last measur
• e weight to the last measurement. By assigning it a weight of 5 we indicated tha
• . Taking the case to the extreme and assigning a weight of 0.0001 to the last me
• attack
• Puzzle 4 It did not look sensible to attack puzzle 4 by hand. Too many chads, ab
• attempt
• tranets and business websites offen attempt to emulate this facility provinding
• attribute
• g does not need to have the "href" attribute as its first attribute. Likewise,
• the "href" attribute as its first attribute. Likewise, although the use of low
• ise, although the use of lowercase attributes such as "a" and "href" is prefere
• which still use uppercase tags and attributes. Finally, searching for "href" it
• and media tags also use the "href" attribute but they do not link to a web page
• attributes
• ise, although the use of lowercase attributes such as "a" and "href" is prefered
• which still use uppercase tags and attributes. Finally, searching for "href" its
• authors
• aving it. Likewise, self-publishing authors like me could benefit of having suc
• automatic
• he filtering process is completely automatic. The function receives as input a
• s with the suggested words happens automatically, after which the user is allow
• automatically
• with the suggested words happens automatically, after which the user is allow t
• autonomous
• l roboticists in many years, whose autonomous cars placed 1st and 2nd in the DAR
• l roboticists in many years, whose autonomous cars placed 1st and 2nd in the DAR
• available
• tion. The output of this function, available for inspection as a file, is the s
• p with me and the hours that I had available to put in the challenge dwindle to



• backwards
• tic of our society. We have it all backwards. Even now, when the consequences o
• band
• n which you could see, on the back, a band stand. The location was the Jose Mart
• beautiful
• chads with red ink, and more. The beautiful part is that all of these approach
• believe
• f the robot, i.e., x 0 , although I believe it could be assigned to any one pos
• to an equation (i.e., W 1) when we believe that that particular equation has a
• belong
• ng words that are very likely to not belong in an index, e.g., pronouns, conjunc
• benefit
• of their website but still sees the benefits of having it. Likewise, self-publi
• lf-publishing authors like me could benefit of having such a program to aid in
• benefits
• of their website but still sees the benefits of having it. Likewise, self-publis
• black
• t the pages visited using white- and black-site lists. For example, to index a
• : white_list = {"mybook", "chapter"} black_list = {"copyrights", "glosary"} whi
• s. First, the user can set white and black lists of entries that specify words
• word list, it would be reasonable to black list "Albert" and white list the com
• ld not. The merging of the white and black lists with the suggested words happe
• black-site
• the pages visited using white- and black-site lists. For example, to index a boo
• black_list
• white_list = {"mybook", "chapter"} black_list = {"copyrights", "glosary"} which
• blank
• swim, or to _____ (you fill out the blank) early is extraordinarily sensible.
• block
• of functions, shown in purple in the block diagram: Crawl_web_by_keyword Sugges
• blue
• int. The setup Distances are shown in blue, weights are shown in red. The constr
• web that are being indexed (shown in blue). The result of the four routines, sh
• boils
• , a release note, etc. The situation boils to this: when you are looking for a
• book
• ite and my essays about the an online book that I am writing. I submitted the pr
• t to create indexes of intra-nets and books-on-line I wanted to create an index
• ntents of my website and of an online books that I am writting, a function that
• creation of indexes for intranets and books-on-line. At the end of any paper-bou
• e. At the end of any paper-bound text book or reference book we find an index wi
• ny paper-bound text book or reference book we find an index with words that tell
• know what to search for. Paper-bound books give a broad-topic overview with a t
• sk of creating an index for an online book, which by its very nature, it is boun
• k-site lists. For example, to index a book we might have: white_list = {"mybook"
• book we might have: white_list = {"mybook", "chapter"} black_list = {"copyright
• ts", "glosary"} which would crawl the book and all its chapters but will not cra
• books
• to create indexes of intra-nets and books-on-line I wanted to create an index
• tents of my website and of an online books that I am writting, a function that
• reation of indexes for intranets and books-on-line. At the end of any paper-bou
• know what to search for. Paper-bound books give a broad-topic overview with a t
• books-on-line
• create indexes of intra-nets and books-on-line I wanted to create an index of t
• ion of indexes for intranets and books-on-line. At the end of any paper-bound t
• borders
• the coffee mug stain, assemble the borders, assemble the chads with red ink, a
• bottom
• xt: 10, 23, 8, 24, 18 and 21. At the bottom of the message, they show you what t
• pieces with the top 20% dark and the bottom 40% white". This looked very useful.
• bound
• oks-on-line. At the end of any paper-bound text book or reference book we find
• o not know what to search for. Paper-bound books give a broad-topic overview wi
• ook, which by its very nature, it is bound to change everytime that a change is
• nt connotation. Also, the program is bound to omit words that the user consider
• brain
• , both natural and artificial, e.g., brain organization, theories of early lear
• breath-first
• the number of pages visited and a breath-first version that limits the pages vis
• brief
• d cs-262 classes. The following is a brief description of the four main functio
• broad
• search for. Paper-bound books give a broad-topic overview with a table of conte
• tion in the table of contents is too broad. Instead, that of an index has the r
• broad-topic
• rch for. Paper-bound books give a broad-topic overview with a table of contents
• browser
• y were developing Mosaic, the first browser and the base code for both Netscape
• built
• himself in a map of the environment built as it explores its surroundings. In
• business
• rd or topic is found. Intranets and business websites offen attempt to emulate t



• calculation
• rs / Udacity Autonomous Car Quick calculation of the and matrices for Graph-SLA
• calibration
• on (e.g., image formation, camera calibration, image enhancement) and mid-level
• camera
• level vision (e.g., image formation, camera calibration, image enhancement) and
• cannot
• er wants to index are disjointed and cannot be crawled from a single parent, the
• canvas
• wing program and thus, using it as a canvas to assemble the chads was slow. It t
• : I could not zoom in and out of the canvas and, since the canvas was limited in
• and out of the canvas and, since the canvas was limited in size, I had to have m
• simultaneous and partially redundant canvases. Still, I had gotten better at usi
• canvases
• imultaneous and partially redundant canvases. Still, I had gotten better at usin
• capable
• mportant because the program is not capable to measure the importance of combin
• cars
• cists in many years, whose autonomous cars placed 1st and 2nd in the DARPA's Gra
• cists in many years, whose autonomous cars placed 1st and 2nd in the DARPA's Gra
• case
• e elements of a row or column. In the case of the initial position, the sum shou
• ations are now consistent. Taking the case to the extreme and assigning a weight
• e. Likewise, although the use of lowercase attributes such as "a" and "href" is
• e are many pages which still use uppercase tags and attributes. Finally, searchi
• add his/her own, for example, for the case of localization. The output of this f
• as the previous one. If that was the case, puzzle 3 should take 40 hrs to solve
• sense to join forces but even in that case it is unlikely that we would have won
• adjecent to a particular one. In this case, each chad constrains all the others
• chad
• his puzzle. This puzzle had about 200 chads, i.e., it was a warm-up. With a late
• sophisticated way possible: I cut the chads from the tiff images in which they c
• using it as a canvas to assemble the chads was slow. It took me about 20 hrs to
• ok me about 20 hrs to assemble enough chads to be able to read the message of th
• zle 2 The second puzzle had about 350 chads but unlike the first one, it had bee
• n, assemble the borders, assemble the chads with red ink, and more. The beautifu
• eared useful. Puzzle 1 had 150 useful chads and it took about 10 hrs to solve (o
• should have). Puzzle 2 had 300 useful chads and took about 20 hrs. Puzzle 3 had
• 20 hrs. Puzzle 3 had about 600 useful chads. It appeared that the puzzles had be
• to attack puzzle 4 by hand. Too many chads, about 2300, most of them useful, an
• onstraints. However, there were a few chads written in a distictive ink color, h
• knownst to the contestants, these few chads contained the answers to the questio
• t together a few of these distinctive chads and got 'Dave' and 'George' in a mat
• hurch. Eventually, I assembled enough chads to recognize the structure in the mi
• you the most likely configurations of chads that are adjecent to a particular on
• a particular one. In this case, each chad constrains all the others and hopeful
• chads
• is puzzle. This puzzle had about 200 chads, i.e., it was a warm-up. With a late
• ophisticated way possible: I cut the chads from the tiff images in which they c
• using it as a canvas to assemble the chads was slow. It took me about 20 hrs to
• k me about 20 hrs to assemble enough chads to be able to read the message of th
• le 2 The second puzzle had about 350 chads but unlike the first one, it had bee
• , assemble the borders, assemble the chads with red ink, and more. The beautifu
• ared useful. Puzzle 1 had 150 useful chads and it took about 10 hrs to solve (o
• hould have). Puzzle 2 had 300 useful chads and took about 20 hrs. Puzzle 3 had
• 0 hrs. Puzzle 3 had about 600 useful chads. It appeared that the puzzles had be
• to attack puzzle 4 by hand. Too many chads, about 2300, most of them useful, an
• nstraints. However, there were a few chads written in a distictive ink color, h
• nownst to the contestants, these few chads contained the answers to the questio
• together a few of these distinctive chads and got 'Dave' and 'George' in a mat
• urch. Eventually, I assembled enough chads to recognize the structure in the mi
• ou the most likely configurations of chads that are adjecent to a particular on
• chain
• bout x 1 = x 0 + d 0 1 We can form a chain of relative motions of this kind, ea
• challenge
• ex puzzles. In 2011, they issued a challenge to put together, in a month, five
• solving the "easy" portion of the challenge? Find out answering the questions
• her contestants, I found about the challenge in slashdot , after it had already
• that I had available to put in the challenge dwindle to nothing marking the end
• changes
• we had to adapt a little. The main changes that we had to make to the crawl_we
• chapter
• ight have: white_list = {"mybook", "chapter"} black_list = {"copyrights", "glos
• ch would crawl the book and all its chapters but will not crawl the pages devot
• chapters
• ch would crawl the book and all its chapters but will not crawl the pages devote
• characters
• not a text. You can narrow posible characters that follow a partial word but thi
• first 3 numbers correspond to the characters "JGO". The last 3 will correspond
• check
• , alternatively, to use the graph to check if the matrices are correct. Backgro
• children
• ntly taking some years off while my children grow up. My wife Becky and I have t
• up. My wife Becky and I have three children: Sofi, now 4 1/2, and twins Danny a
• choice
• is seldom used. Instead, the tool of choice of a website to guide a user to its
• it looked as if I had made the right choice. After 40 hrs I had assembled about
• church
• ought that the drawing referred to a church. Eventually, I assembled enough chad
• cipher
• on for this puzzle was What is the deciphered message? Not so easy, right? Give
• nswer. Puzzle 2 4 pts: What is the deciphered message? I overthought this questi
• e of the message, I thought that the cipher was a simple substitution Caesar cip
• her was a simple substitution Caesar cipher. However, in spite that the Caesar c
• r. However, in spite that the Caesar cipher is a very simple one, the encipherin
• r cipher is a very simple one, the enciphering in the message was even simpler.
• ill correspond to "MEZ" Hence, the deciphered message is "JGOMEZ" Puzzle 3 2 pts
• city
• try? (easy) and What is the indicated city/town? (hard). Can you figure them out
• country? 6 pts: What is the indicated city/town? The country is easy enough to s
• sy enough to spot: Cuba. However, the city/town is very difficult to figure out.
• together, were the coordinates of the city. I turns out that they were the coord
• ou that the map was that of Cuba. The city has to be deduced exclusively from th
• city/town
• ? (easy) and What is the indicated city/town? (hard). Can you figure them out f
• ntry? 6 pts: What is the indicated city/town? The country is easy enough to spo
• enough to spot: Cuba. However, the city/town is very difficult to figure out. F
• class
• Index Maker I took Udacity 's CS-101 class - Building a Search Engine - in Marc
• in March 2012. Initially I took this class because I wanted to take CS-373 but
• orked sounded to alluring to me. The class was a surprise: fast paced, interest
• to use. Based on the contents of the class I wrote a program that could index m
• to the open contest that closed the class. Udacity - CS-373 - Autonomous car -
• ost on SLAM I took Udacity 's CS-373 class - Programming a Robotic Car - in Mar
• g a Robotic Car - in March 2012. The class was taught by Sebastian Thrun, one o
• . Background I took Udacity's CS-373 class - Programming a Robotic Car - in Mar
• g a Robotic Car - in March 2012. The class was taught by Sebastian Thrun, one o
• e solution This example was given in class but, although the derivation of the
• o after taking the cs-253 and cs-262 classes. The following is a brief descript
• combination of words. As seen in the class, "monty" and "python" might be relev
• classes
• after taking the cs-253 and cs-262 classes. The following is a brief descripti
• click
• . I am giving you the puzzle solved (click on the image). What would have been
• closer
• er. In addition, it is fun, gets you closer to your kids, prepares them for the
• clue
• ng in the message, "Comrade", was the clue that "Tovarich" was the correct answe
• tes of La Havana and Nassau, meant to clue you that the map was that of Cuba. Th
• code
• osaic, the first browser and the base code for both Netscape IE. I had a web pag
• unctions seen in the course. Download code Examples: index of this website index
• coffee
• paper, with two ink colors and had a coffee mug mark, i.e., a lot of hints. Henc
• o approach this puzzle: assemble the coffee mug stain, assemble the borders, ass
• collaborator
• e the name and/or initials of the collaborators? Presumably, the "collaborators"
• e collaborators? Presumably, the "collaborators" were the people that had writte
• f a few hours. I also found a 3rd collaborator but I did not realize it, so I di
• e the name and/or initials of the collaborators? I got 3 pts for 'George' and 'D
• collaborators
• the name and/or initials of the collaborators? Presumably, the "collaborators"
• collaborators? Presumably, the "collaborators" were the people that had writte
• the name and/or initials of the collaborators? I got 3 pts for 'George' and 'D
• colors
• written in lined-paper, with two ink colors and had a coffee mug mark, i.e., a l
• column
• to add all the elements of a row or column. In the case of the initial position
• combination
• able to measure the importance of combination of words. As seen in the class, "
• relevant but "monty python", as a combination, has a different connotation. Als
• of entries that specify words or combination of words that the user definitely
• list "Albert" and white list the combination "Albert Einstein". In this way bo
• e list and add or remove words or combinations of words on the fly. The output
• combinations
• e list and add or remove words or combinations of words on the fly. The output o
• commercial
• online documents: HTML indexer , a commercial package that runs on Windows, and
• completely
• 00 words. The filtering process is completely automatic. The function receives a
• he puzzle without the tool, it was completely feasible to answer the questions w
• complex
• d in understanding how people solve complex puzzles. In 2011, they issued a cha
• d in understanding how people solve complex puzzles. They issue the Shredder Ch
• computatinal
• en I would need to put together a computatinal tool to help me assemble the puzz
• computational
• rs to put it together. Indeed, a computational tool was not needed for this puz
• 1 and 2 definitely did no need a computational tool to solve them. Puzzles 4 an
• Instead I started to work on the computational tool that seemed the only viable
• anking teams, had put together a computational tool that help them select piece
• computer
• Engineering I work in robotics and computer vision. In robotics, I have worked
• concordance
• ype of index commonly known as a "concordance", i.e., a very specific type of i
• This function sieves the general concordance index produced by the web crawler
• y a factor of 2-to-3, i.e, if the concordance had 3000 words, the index of sugg
• n as a file, is the subset of the concordance whose entries were deemed likely
• confidence
• urement. Finally, depending on our confidence on a particular constraint we can
• configuration
• t could give you the most likely configurations of chads that are adjecent to a
• opefully settles on a reasonable configuration. However, I did not finish my to
• configurations
• t could give you the most likely configurations of chads that are adjecent to a
• conjunctions
• long in an index, e.g., pronouns, conjunctions, prepositions, contractions, comm
• connect
• note and more time pondering how to connect them. There were many ways to appro
• connotation
• as a combination, has a different connotation. Also, the program is bound to om
• consequence
• nificant errors. This result is a consequence of assigning a large weight to th
• consequences
• all backwards. Even now, when the consequences of our actions no longer need to
• considers
• bound to omit words that the user considers important and to add words that th
• consistent
• e and all the observations are now consistent. Taking the case to the extreme an
• constrains
• lue, weights are shown in red. The constrains shown on the graph are: Initial po
• cular one. In this case, each chad constrains all the others and hopefully settl
• constraint
• scribes the motion and measurement constraints or, alternatively, to use the gra
• matrices and encode three type of constraints: Absolute Position Relative Posit
• is given by the Absolute Position constraint, commonly assigned to the initial
• rent locations x i lead to similar constraints. A landmark L j observed from pos
• on our confidence on a particular constraint we can add weight to an equation,
• he vector encodes the value of the constraint while the matrix encodes the locat
• locations that are involved in the constraint. Once the and matrices are set up,
• ion set with the absolute position constraint. The setup Distances are shown in
• the weight of the initial position constraint. In the example, when adding the e
• 3 - 1 + 0 - 1 = 1 For every other constraint, either relative pose or landmark,
• most of them useful, and very few constraints. However, there were a few chads
• n mind: an automated tool based on constraint satisfaction. I was not expecting
• constraints
• cribes the motion and measurement constraints or, alternatively, to use the gra
• matrices and encode three type of constraints: Absolute Position Relative Posit
• ent locations x i lead to similar constraints. A landmark L j observed from pos
• most of them useful, and very few constraints. However, there were a few chads
• contact
• a, CA 91109 my JPL web page You can contact me at andres@acastano.com Hi.. I am
• contains
• a very specific type of index that contains every single string found. Although
• content
• answers stood out from the general content of the puzzle. Hence, I did not try
• contents
• he opportunity to use. Based on the contents of the class I wrote a program that
• I wanted to create an index of the contents of my website and of an online book
• is that unless you already know the contents of the website you do not know what
• road-topic overview with a table of contents. This facility is not often offered
• nce you do not look at the table of contents; you look at the index. The informa
• ex. The information in the table of contents is too broad. Instead, that of an i
• contest
• ge describes my experience with the contest. Udacity - CS-101 - Search Engine -
• the program as an entry to the open contest that closed the class. Udacity - CS
• index of CS-101 forum entry for the contest What is Index Maker Index Maker is
• ey issue the Shredder Challenge , a contest to put together, from Oct. 27 to De
• y results. Puzzle 1 Like many other contestants, I found about the challenge in
• that stood out. Unbeknownst to the contestants, these few chads contained the
• the end of my participation in the contest. I never got to give a try to puzzl
• the Cienfuegos City Hall. After the contest I found out that the drawing was ba
• espectively. Departing thoughts The contest was a lot of fun and very challengi
• ave been part of a team. During the contest I was in touch with another team, t
• Dolinar. However, they entered the contest even later than me and thus when th
• contestants
• results. Puzzle 1 Like many other contestants, I found about the challenge in s
• hat stood out. Unbeknownst to the contestants, these few chads contained the an
• contractions
• ouns, conjunctions, prepositions, contractions, common verbs, etc. The user can
• coordinates
• or a long time I thought that the coordinates in the message, which I had not b
• en able to put together, were the coordinates of the city. I turns out that the
• y. I turns out that they were the coordinates of La Havana and Nassau, meant to
• coordination
• nfigurable robots and hand-to-eye coordination, using manipulators. In vision, I
• copious
• e activities carried out during the copious free time left from being a full-ti
• copyright
• mybook", "chapter"} black_list = {"copyrights", "glosary"} which would crawl th
• not crawl the pages devoted to the copyright or the glosary. Memoization of all
• copyrights
• mybook", "chapter"} black_list = {"copyrights", "glosary"} which would crawl the
• corporation
• does not want to approach a large corporation like Google or Yahoo to create a
• correspond
• he last three. The first 3 numbers correspond to the characters "JGO". The last
• characters "JGO". The last 3 will correspond to "MEZ" Hence, the deciphered mes
• corresponds
• The value of each element of that corresponds to a node is the sum of the weigh
• count
• mall error (low variance) and should count more towards finding the solution of
• country
• this puzzle: What is the indicated country? (easy) and What is the indicated c
• zzle 3 2 pts: What is the indicated country? 6 pts: What is the indicated city/
• hat is the indicated city/town? The country is easy enough to spot: Cuba. Howev
• course
• , respectively. The students of this course had access to a very active forum th
• robably will become private once the course is finished. Here I am reproducing a
• e, respectively. The student of this course had access to a very active forum. H
• this example taken straight from the course, the robot is moving along a line, o
• search engine functions seen in the course. Download code Examples: index of th
• how to set up such interface in the course and neither my knowledge of Python n
• crawl
• cription of the four main functions. crawl_web_by_keyword The starting point fo
• arting point for this routine is the crawl_web function from Udacious. We saw t
• three versions of it: indiscriminate crawling, a version that limits the number
• hose that fall within a level of the crawl. Unfortunatelly, none of this versio
• n changes that we had to make to the crawl_web routine are: Restrict the pages
• "copyrights", "glosary"} which would crawl the book and all its chapters but wi
• ok and all its chapters but will not crawl the pages devoted to the copyright o
• ll the visited pages to speed up the crawling and the functions that will creat
• o index are disjointed and cannot be crawled from a single parent, the user can
• oncordance index produced by the web crawler to produce a small set of words th
• crawl_web
• iption of the four main functions. crawl_web_by_keyword The starting point for
• ting point for this routine is the crawl_web function from Udacious. We saw thr
• changes that we had to make to the crawl_web routine are: Restrict the pages vi
• crawl_web_by_keyword
• n of the four main functions. crawl_web_by_keyword The starting point for this r
• crawler
• ncordance index produced by the web crawler to produce a small set of words tha
• create
• rch Engine Index Maker - a script to create indexes of intra-nets and books-on-l
• a-nets and books-on-line I wanted to create an index of the contents of my websi
• to have. There are only two tools to create indexes from online documents: HTML
• corporation like Google or Yahoo to create a simple index of their website but
• crawling and the functions that will create the index. Deal with the retrieval o
• ds that will be in the online index. create_html_index This final function makes
• create_html_index
• t will be in the online index. create_html_index This final function makes a we
• creation
• thon script that semi-automates the creation of indexes for intranets and books-
• credit
• bmit it as an answer and did not get credit for it. Can you spot it? Likewise, I
• critic
• ing, etc. Social issues I am an acid critic of our society. We have it all backw



• data
• ven got to learn about Hash tables, a data structure that I had not have the opp
• for: a driver, a document, a piece of data, a release note, etc. The situation b
• datum
• to this: when you are looking for a datum in a reference you do not look at th
• days
• nually. It is my tribute to the early days of the web. While I was working with
• default
• web page is ready to put online. By default, it is a strict XHTML 1.0 page in t
• derivation
• s given in class but, although the derivation of the equations is correct, the s
• derive
• SLAM The following is a quick way to derive the and matrices from the graph that
• describes
• ing the name 'Goldsong'. This page describes my experience with the contest. Ud
• e and matrices from the graph that describes the motion and measurement constra
• description
• classes. The following is a brief description of the four main functions. crawl
• destination
• a website to guide a user to its destination is its tree structure, which ofte
• detection
• and mid-level vision (e.g., object detection, recognition and identification).
• diagonal
• trix, we could start by finding its diagonal. Each element of the diagonal is th
• g its diagonal. Each element of the diagonal is the sum of the weights of the in
• ular node: Then we can find the off-diagonal elements: for each edge i, j it is
• diagram
• rytime that a change is made. Block diagram of Index Maker Index Maker is compo
• tions, shown in purple in the block diagram: Crawl_web_by_keyword Suggest_index
• different
• observation of the landmarks from different locations x i lead to similar cons
• y python", as a combination, has a different connotation. Also, the program is
• s surprised that their tool was so different from what I had in mind: an automa
• difficult
• ments, each one progressively more difficult. Points were awarded, not for solv
• ents , each one progressively more difficult than the previous one. There were
• le. And each time it was ever more difficult to put pieces together. Unlike puz
• ba. However, the city/town is very difficult to figure out. For a long time I t
• direction
• l tool that seemed the only viable direction to solve puzzle 4. Big mistake! Al
• discuss
• n your answer to the question? I'll discuss my results at the end of the page.
• dissabilities
• allows you to identify learning dissabilities that otherwise would not have be
• distance
• L j observed from position x i at a distance k is described as L j = x i + k in
• which it observes the landmark at a distance of +10. This means that the landmar
• distant
• at I could have ranked is 7th, very distant from the score of 50 pts. earned by
• distictive
• here were a few chads written in a distictive ink color, hand writing or orienta
• distinctive
• re. I put together a few of these distinctive chads and got 'Dave' and 'George'
• ot try to assemble the words with distinctive hand writing or orientation. Inst
• doable
• should take 40 hrs to solve, pretty doable. Thus, I had my answer and Boy, was
• document
• at s/he is looking for: a driver, a document, a piece of data, a release note, e
• tools to create indexes from online documents: HTML indexer , a commercial packa
• documents
• ogether, in a month, five shredded documents, each one progressively more diffi
• ools to create indexes from online documents: HTML indexer , a commercial packa
• . 27 to Dec 2, 2011, five shredded documents , each one progressively more diff
• domain
• useful to travel exclusively a given domain so we had to adapt a little. The mai
• dome
• c search for the park showed that the dome was that of the Cienfuegos City Hall.
• drawn
• only one ink color, not two, and was drawn on white paper, not lined one. Howev
• driver
• e stores what s/he is looking for: a driver, a document, a piece of data, a rele
• dwindle
• d available to put in the challenge dwindle to nothing marking the end of my pa



• easier
• rdinarily sensible. As a rule, it is easier to teach anything to a 2-year old th
• easily
• Java. I think that a new program can easily make an impact in the area of indexi
• easy
• us we have the Omega matrix: Also, an easy way to verify that the matrix is corr
• lenge How do you measure solving the "easy" portion of the challenge? Find out a
• f hints. Hence, it looked also pretty easy to do by hand. The time will come whe
• hat is the deciphered message? Not so easy, right? Give it a try. Remember that
• follow a partial word but this is not easy with a drawing. Likewise, the drawing
• zzle: What is the indicated country? (easy) and What is the indicated city/town?
• e indicated city/town? The country is easy enough to spot: Cuba. However, the ci
• edge
• weights of the incoming and outgoing edges to the particular node: Then we can
• d the off-diagonal elements: for each edge i, j it is the negative of the weight
• e is the sum of the weighted incoming edges to the node minus the sum of the wei
• inus the sum of the weighted outgoing edges: Comment on the solution This exampl
• edges
• weights of the incoming and outgoing edges to the particular node: Then we can
• is the sum of the weighted incoming edges to the node minus the sum of the wei
• nus the sum of the weighted outgoing edges: Comment on the solution This exampl
• element
• start by finding its diagonal. Each element of the diagonal is the sum of the w
• : Then we can find the off-diagonal elements: for each edge i, j it is the nega
• drawing the graph is to add all the elements of a row or column. In the case of
• nt. In the example, when adding the elements of the x 0 row in we get 3 - 1 + 0
• 0. In the example, when adding the elements of the row of x 2 in we get 0 - 1
• 5 = 0 The vector The value of each element of that corresponds to a node is th
• elements
• : Then we can find the off-diagonal elements: for each edge i, j it is the negat
• drawing the graph is to add all the elements of a row or column. In the case of
• nt. In the example, when adding the elements of the x 0 row in we get 3 - 1 + 0
• 0. In the example, when adding the elements of the row of x 2 in we get 0 - 1 +
• emulate
• business websites offen attempt to emulate this facility provinding a local se
• encode
• near equations = X. The matrices and encode three type of constraints: Absolute
• ned a low weight (0 W 1). The vector encodes the value of the constraint while t
• e of the constraint while the matrix encodes the locations that are involved in
• encodes
• ed a low weight (0 W 1). The vector encodes the value of the constraint while t
• of the constraint while the matrix encodes the locations that are involved in
• engine
• topic of understanding how a search engine worked sounded to alluring to me. Th
• ex maker program based on the search engine functions seen in the course. Downlo
• s facility provinding a local search engines. The problem is that unless you alr
• ngle string found. Although a search engine will return a valid result to an arb
• engineer
• at andres@acastano.com Hi.. I am an engineer, presently taking some years off wh
• engines
• facility provinding a local search engines. The problem is that unless you alr
• english
• t is a strict XHTML 1.0 page in the english language that could pass any valida
• enhancement
• mation, camera calibration, image enhancement) and mid-level vision (e.g., obje
• entries
• Crawl_web_by_keyword Suggest_index_entries select_index_entries Create_html_in
• Suggest_index_entries select_index_entries Create_html_index The input to thes
• to find in an index. suggest_index_entries This function sieves the general co
• the subset of the concordance whose entries were deemed likely to be relevant t
• specific information. select_index_entries This function uses the index of sug
• unction uses the index of suggested entries as the starting point to obtain a f
• ows the user to alter the suggested entries in two ways. First, the user can se
• er can set white and black lists of entries that specify words or combination o
• t Einstein" and "Einstein" would be entries but "Albert", by itself, would not.
• n the previous one. There were 9000 entries, 69 of which scored at least 1 of 5
• entry
• iting. I submitted the program as an entry to the open contest that closed the
• f this website index of CS-101 forum entry for the contest What is Index Maker
• environment
• to locate himself in a map of the environment built as it explores its surround
• equally
• that all of these approaches worked equally well. In the end, it took me 20 hrs
• equation
• ents are written as a set of linear equations = X. The matrices and encode three
• constraint we can add weight to an equation, leading to a solution in which the
• which the error for that particular equation is reduced more than the others, e.
• + W*k We assign a large weigh to an equation (i.e., W 1) when we believe that th
• hen we believe that that particular equation has a small error (low variance) an
• olution of the system. Likewise, an equation associated with a large error (high
• but, although the derivation of the equations is correct, the setup of the equat
• ations is correct, the setup of the equations had an error. Let's see: Solving =
• t the measurement was lousy and the equation should be given no value when solvi
• equations
• nts are written as a set of linear equations = X. The matrices and encode three
• ut, although the derivation of the equations is correct, the setup of the equat
• tions is correct, the setup of the equations had an error. Let's see: Solving =
• error
• ute. The problem is that there is an error associated with the motion and thus,
• only an approximation because of the errors in both the location of the robot a
• , leading to a solution in which the error for that particular equation is redu
• that particular equation has a small error (low variance) and should count more
• an equation associated with a large error (high variance) should be assigned a
• t, the setup of the equations had an error. Let's see: Solving = X gives us X =
• +3 as commanded but made significant errors. This result is a consequence of as
• obot moved as commanded with smaller errors than before and all the observation
• ] T i.e., the expected result for an error-free scenario. This work is licensed
• error-free
• T i.e., the expected result for an error-free scenario. This work is licensed un
• errors
• only an approximation because of the errors in both the location of the robot an
• +3 as commanded but made significant errors. This result is a consequence of ass
• obot moved as commanded with smaller errors than before and all the observations
• essays
• m that could index my website and my essays about the an online book that I am w
• event
• ey do not link to a web page. In the event that the pages that the user wants t
• everytime
• very nature, it is bound to change everytime that a change is made. Block diagr
• example
• ent: L = x 2 +1. Weight = 5 In this example taken straight from the course, the
• initial position constraint. In the example, when adding the elements of the x
• ndmark, the sum should be 0. In the example, when adding the elements of the ro
• edges: Comment on the solution This example was given in class but, although th
• ng white- and black-site lists. For example, to index a book we might have: whi
• hich " a href=" in insuficient. For example, an anchor tag does not need to hav
• ese filters or add his/her own, for example, for the case of localization. The
• o have or to omit respectively. For example, if the words "Albert" and "Einstei
• exclusively
• this versions is useful to travel exclusively a given domain so we had to adapt
• Cuba. The city has to be deduced exclusively from the drawing. Initially I tho
• expect
• not the type of information that we expect to find in an index. suggest_index_e
• experience
• 'Goldsong'. This page describes my experience with the contest. Udacity - CS-101
• lace 10th. This is a summary of my experience with the puzzles and my results. P
• explain
• nswer an associated question and to explain how you figured out the answer. Per
• explores
• map of the environment built as it explores its surroundings. In Graph-SLAM, th
• extraordinarily
• ou fill out the blank) early is extraordinarily sensible. As a rule, it is easi
• extreme
• consistent. Taking the case to the extreme and assigning a weight of 0.0001 to



• facility
• sites offen attempt to emulate this facility provinding a local search engines.
• view with a table of contents. This facility is not often offered by a website a
• factor
• tion in the size of the indexes by a factor of 2-to-3, i.e, if the concordance h
• family
• Cosita, our Shih-Tzu, rounds up the family. Family matters We wanted our kids t
• feasible
• without the tool, it was completely feasible to answer the questions without ass
• figure
• ated question and to explain how you figured out the answer. Personally, I found
• indicated city/town? (hard). Can you figure them out from the partially-assemble
• , the city/town is very difficult to figure out. For a long time I thought that
• file
• nput to these functions are parameter files that can be modified by the user (sh
• he user (shown in green) and the html files on the web that are being indexed (s
• our routines, shown in red, is a HTML file ready to be uploaded to the website.
• ould have had access to the parameter files via a GUI or, even better, a wep pag
• unction receives as input a number of files containing words that are very likel
• nction, available for inspection as a file, is the subset of the concordance who
• altering the page by modifying a CSS file. This work is licensed under a Creati
• files
• put to these functions are parameter files that can be modified by the user (sh
• e user (shown in green) and the html files on the web that are being indexed (s
• uld have had access to the parameter files via a GUI or, even better, a wep pag
• nction receives as input a number of files containing words that are very likel
• filters
• t to use all, some or none of these filters or add his/her own, for example, fo
• final
• es as the starting point to obtain a final index. This step is important becaus
• online index. create_html_index This final function makes a web page using the
• finish
• tually they did catch up with me and finished in 13th place. In retrospective, i
• le configuration. However, I did not finish my tool and words are wind.
• floor
• working with my robots in the first floor of the Beckman Institute in Urbana,
• e Beckman Institute in Urbana, three floors above, in NCSA, they were developin
• floors
• e Beckman Institute in Urbana, three floors above, in NCSA, they were developing
• focus
• s no longer need to be predicted, we focus on our short-term personal issues in
• forces
• ve, it would have made sense to join forces but even in that case it is unlikely
• form
• x 1 is about x 1 = x 0 + d 0 1 We can form a chain of relative motions of this k
• formation
• k in low-level vision (e.g., image formation, camera calibration, image enhance
• forum
• s course had access to a very active forum that probably will become private on
• Here I am reproducing a post in the forum about SLAM that many people found in
• s course had access to a very active forum. Here I reproduce a post about SLAM
• ndex of this website index of CS-101 forum entry for the contest What is Index
• freeware
• hat runs on Windows, and XRefHT , a freeware with versions for both Windows and
• full-time
• opious free time left from being a full-time dad of three toddlers. This page i
• function
• online books that I am writting, a function that most word processors do not of
• program based on the search engine functions seen in the course. Download code
• ly to be useful to you. That is the functionality that we want to have. There ar
• ex Maker Index Maker is composed of functions, shown in purple in the block diag
• reate_html_index The input to these functions are parameter files that can be mo
• brief description of the four main functions. crawl_web_by_keyword The starting
• t for this routine is the crawl_web function from Udacious. We saw three version
• es to speed up the crawling and the functions that will create the index. Deal w
• d to be indexed. The output of this function is a type of index commonly known a
• n index. suggest_index_entries This function sieves the general concordance inde
• rocess is completely automatic. The function receives as input a number of files
• of localization. The output of this function, available for inspection as a file
• ormation. select_index_entries This function uses the index of suggested entries
• hat the user does not care for. The function allows the user to alter the sugges
• ords on the fly. The output of this function is an list with all the words that
• index. create_html_index This final function makes a web page using the list of
• functionality
• to be useful to you. That is the functionality that we want to have. There are
• functions
• program based on the search engine functions seen in the course. Download code
• x Maker Index Maker is composed of functions, shown in purple in the block diag
• eate_html_index The input to these functions are parameter files that can be mo
• brief description of the four main functions. crawl_web_by_keyword The starting
• s to speed up the crawling and the functions that will create the index. Deal w
• future
• to your kids, prepares them for the future, and allows you to identify learning



• general
• on and identification). However, in general, I am interested in anything relate
• ex_entries This function sieves the general concordance index produced by the w
• all the answers stood out from the general content of the puzzle. Hence, I did
• glosary
• pter"} black_list = {"copyrights", "glosary"} which would crawl the book and al
• ges devoted to the copyright or the glosary. Memoization of all the visited pag
• grade
• pass any validation test with a good grade. The appereance of the page can be c
• granularity
• d, that of an index has the right granularity to be the most likely to be usefu
• graph
• to derive the and matrices from the graph that describes the motion and measur
• raints or, alternatively, to use the graph to check if the matrices are correct
• in red. The constrains shown on the graph are: Initial position: x 0 = -3 Rela
• atrix is correct without drawing the graph is to add all the elements of a row
• green
• an be modified by the user (shown in green) and the html files on the web that
• grow
• king some years off while my children grow up. My wife Becky and I have three ch
• guide
• , the tool of choice of a website to guide a user to its destination is its tre



• hand
• ereance of the page can be changed by hand, modifying the html page or, without
• e puzzle, I had no qualms doing it by hand, in the most unsophisticated way poss
• , it looked also pretty easy to do by hand. The time will come when I would need
• was an unknown. Should I solve it by hand or start to write the tool? Well.. th
• t look sensible to attack puzzle 4 by hand. Too many chads, about 2300, most of
• ds written in a distictive ink color, hand writing or orientation that stood out
• o assemble the words with distinctive hand writing or orientation. Instead I sta
• hand-to-eye
• ed with reconfigurable robots and hand-to-eye coordination, using manipulators.
• happens
• lack lists with the suggested words happens automatically, after which the user
• hellish
• h a program to aid in the otherwise hellish task of creating an index for an on
• help
• o put together a computatinal tool to help me assemble the puzzle but this was n
• ut together a computational tool that help them select pieces. You could ask the
• highest
• ional answer in puzzle 4. Thus, the highest that I could have ranked is 7th, ve
• highly
• ur of its six members and think very highly of them: Mike Burl, Aaron Kiely, Mat
• hindsigh
• he Apple version of Power Point. In hindsigh, my selection of tools was poor. Ke
• hints
• ad a coffee mug mark, i.e., a lot of hints. Hence, it looked also pretty easy t
• hope
• that many people found useful, in the hope that new CS-373 Udacity students also
• hopefully
• chad constrains all the others and hopefully settles on a reasonable configurat
• hours
• e' and 'George' in a matter of a few hours. I also found a 3rd collaborator but
• elly, life caught up with me and the hours that I had available to put in the c
• href
• the retrieval of pages for which " a href=" in insuficient. For example, an anc
• anchor tag does not need to have the "href" attribute as its first attribute. Li
• lowercase attributes such as "a" and "href" is prefered, there are many pages wh
• d attributes. Finally, searching for "href" itself leads to problems because mai
• ail tags and media tags also use the "href" attribute but they do not link to a
• html
• x_entries select_index_entries Create_html_index The input to these functions ar
• by the user (shown in green) and the html files on the web that are being index
• nd neither my knowledge of Python nor html is enough to write such interface. Ho
• t will be in the online index. create_html_index This final function makes a web
• can be changed by hand, modifying the html page or, without altering the page by



• idea
• 80% of the puzzle. However, I had no idea how to answer the questions associate
• identification
• bject detection, recognition and identification). However, in general, I am inte
• identify
• m for the future, and allows you to identify learning dissabilities that otherwi
• image
• n, I work in low-level vision (e.g., image formation, camera calibration, image
• image formation, camera calibration, image enhancement) and mid-level vision (e
• arning, natural language processing, image understanding, etc. Social issues I
• sible: I cut the chads from the tiff images in which they came and pasted them
• you the puzzle solved (click on the image). What would have been your answer t
• ing. A pedestal implies a statue. An image search for statues in Cuba gave me o
• ut that the drawing was based on the image of the park from Wikipedia. Puzzle 4
• images
• sible: I cut the chads from the tiff images in which they came and pasted them i
• impact
• hat a new program can easily make an impact in the area of indexing. The target
• implies
• n it was smooth sailing. A pedestal implies a statue. An image search for statu
• importance
• gram is not capable to measure the importance of combination of words. As seen i
• important
• obtain a final index. This step is important because the program is not capable
• omit words that the user considers important and to add words that the user doe
• in-utero
• lar with Bella who, injured with an in-utero stroke, requires a lot of stimulati
• indexer
• indexes from online documents: HTML indexer , a commercial package that runs on
• indexes
• ne Index Maker - a script to create indexes of intra-nets and books-on-line I w
• that semi-automates the creation of indexes for intranets and books-on-line. At
• There are only two tools to create indexes from online documents: HTML indexer
• have a reduction in the size of the indexes by a factor of 2-to-3, i.e, if the
• indiscriminate
• us. We saw three versions of it: indiscriminate crawling, a version that limits
• information
• tents; you look at the index. The information in the table of contents is too b
• s", "at", this it not the type of information that we expect to find in an inde
• nt to a user looking for specific information. select_index_entries This functi
• initial
• onstraint, commonly assigned to the initial position of the robot, i.e., x 0 ,
• a row or column. In the case of the initial position, the sum should add up to
• should add up to the weight of the initial position constraint. In the example
• initials
• he puzzle: What are the name and/or initials of the collaborators? Presumably, t
• 4 12 pts: What are the name and/or initials of the collaborators? I got 3 pts f
• input
• _index_entries Create_html_index The input to these functions are parameter fil
• automatic. The function receives as input a number of files containing words t
• inspection
• ut of this function, available for inspection as a file, is the subset of the co
• insuficient
• of pages for which " a href=" in insuficient. For example, an anchor tag does
• intelligence
• interested in anything related to intelligence, both natural and artificial, e.g
• interact
• y, after which the user is allow to interact with the list and add or remove wor
• interface
• e did not learn how to set up such interface in the course and neither my knowl
• n nor html is enough to write such interface. Hopefully, I will be able to afte
• intervention
• il much later in life, when early intervention is no longer possible. Science En
• intra-nets
• er - a script to create indexes of intra-nets and books-on-line I wanted to crea
• intranets
• omates the creation of indexes for intranets and books-on-line. At the end of a
• issue
• w people solve complex puzzles. They issue the Shredder Challenge , a contest t
• issues
• ng, image understanding, etc. Social issues I am an acid critic of our society.
• we focus on our short-term personal issues instead of the long-term common ones



• join
• han me and thus when they asked me to join them I was already in 5th place and d
• spective, it would have made sense to join forces but even in that case it is un



• kids
• family. Family matters We wanted our kids to learn as much as possible as early
• timulation. We find that teaching our kids to read, to swim, or to _____ (you fi
• n, it is fun, gets you closer to your kids, prepares them for the future, and al
• knowledge
• rface in the course and neither my knowledge of Python nor html is enough to wr
• known
• e Measurement of a landmark The only known position of the robot is given by th
• function is a type of index commonly known as a "concordance", i.e., a very spe



• landmark
• bot to use repeated observations of landmarks to locate himself in a map of the
• Position Relative Measurement of a landmark The only known position of the robo
• i + d i i+1 The observation of the landmarks from different locations x i lead
• x i lead to similar constraints. A landmark L j observed from position x i at a
• t followed, and the location of the landmarks, all with respect to the location
• s moving along a line, on which the landmark L is also located. The robot moves
• at x=-3, from which it observes the landmark at a distance of +10. This means th
• istance of +10. This means that the landmark is close to -3+10=7. Then the robot
• moves +5 from which it observes the landmark at +5. This means that the robot sh
• should be close to -3+5= +2 and the landmark should be close to 2+5=7, which mat
• moves +3 from which it observes the landmark at +1. This means that the robot is
• obot is at location 2+3 = 5 and the landmark should be close to 5+1=6, which doe
• constraint, either relative pose or landmark, the sum should be 0. In the exampl
• 0, 2.179, 5.714, 6.821] T i.e., the landmark is at 6.821 instead of 7, like the
• landmarks
• ot to use repeated observations of landmarks to locate himself in a map of the
• i + d i i+1 The observation of the landmarks from different locations x i lead
• followed, and the location of the landmarks, all with respect to the location
• language
• theories of early learning, natural language processing, image understanding, et
• trict XHTML 1.0 page in the english language that could pass any validation test
• later
• uld not have been noticed until much later in life, when early intervention is
• wever, they entered the contest even later than me and thus when they asked me
• leaderboard
• d. At this time, I was 5th in the leaderboard. Unfortunatelly, life caught up w
• leads
• Finally, searching for "href" itself leads to problems because mail tags and me
• least
• 9000 entries, 69 of which scored at least 1 of 50 possible points. I participa
• it took about 10 hrs to solve (or at least it should have). Puzzle 2 had 300 us
• level
• visited to those that fall within a level of the crawl. Unfortunatelly, none o
• life
• have been noticed until much later in life, when early intervention is no longer
• h in the leaderboard. Unfortunatelly, life caught up with me and the hours that
• limits
• iscriminate crawling, a version that limits the number of pages visited and a br
• ited and a breath-first version that limits the pages visited to those that fall
• linear
• measurements are written as a set of linear equations = X. The matrices and enco
• lined-paper
• first one, it had been written in lined-paper, with two ink colors and had a co
• link
• the "href" attribute but they do not link to a web page. In the event that the
• lists
• visited using white- and black-site lists. For example, to index a book we mig
• st, the user can set white and black lists of entries that specify words or com
• . The merging of the white and black lists with the suggested words happens aut
• local
• o emulate this facility provinding a local search engines. The problem is that
• er own, for example, for the case of localization. The output of this function,
• localization
• own, for example, for the case of localization. The output of this function, ava
• locate
• epeated observations of landmarks to locate himself in a map of the environment
• ine, on which the landmark L is also located. The robot moves twice, taking a me
• location
• ppose that the robot moves to a new location by an amount d 0 1 . Since x 0 is a
• reference, this would mean that the location of x 1 is also absolute. The proble
• ion of the landmarks from different locations x i lead to similar constraints. A
• n because of the errors in both the location of the robot and the measurement. F
• traint while the matrix encodes the locations that are involved in the constrain
• the vector X to obtain the absolute locations of the robot, i.e., the actual pat
• tual path that it followed, and the location of the landmarks, all with respect
• landmarks, all with respect to the location set with the absolute position cons
• ce, taking a measurement from every location: It starts at x=-3, from which it o
• ches the observation from the first location. Finally, the robot moves +3 from w
• +1. This means that the robot is at location 2+3 = 5 and the landmark should be
• the observations from the first two locations. This last measurement, though, ha
• ersing the tree, trying to find the location where the website stores what s/he
• see, on the back, a band stand. The location was the Jose Marti park in Cienfueg
• locations
• on of the landmarks from different locations x i lead to similar constraints. A
• raint while the matrix encodes the locations that are involved in the constrain
• he vector X to obtain the absolute locations of the robot, i.e., the actual pat
• he observations from the first two locations. This last measurement, though, ha
• long-term
• erm personal issues instead of the long-term common ones. Others Some activitie
• longer
• life, when early intervention is no longer possible. Science Engineering I work
• n the consequences of our actions no longer need to be predicted, we focus on ou
• lousy
• indicating that the measurement was lousy and the equation should be given no
• low-level
• manipulators. In vision, I work in low-level vision (e.g., image formation, cam
• lowercase
• ute. Likewise, although the use of lowercase attributes such as "a" and "href"



• mail
• ref" itself leads to problems because mail tags and media tags also use the "hre
• maker
• fter taking CS-101, I wrote an index maker program based on the search engine f
• manipulators
• d hand-to-eye coordination, using manipulators. In vision, I work in low-level v
• manual
• s without assembling it all, using a manual method. At this time, I was 5th in t
• manually
• s page is pretty plain, all written manually. It is my tribute to the early days
• mark
• h two ink colors and had a coffee mug mark, i.e., a lot of hints. Hence, it look
• t in the challenge dwindle to nothing marking the end of my participation in the
• match
• mark should be close to 2+5=7, which matches the observation from the first loc
• ld be close to 5+1=6, which does not match the observations from the first two
• matches
• ark should be close to 2+5=7, which matches the observation from the first loca
• matrices
• us Car Quick calculation of the and matrices for Graph-SLAM The following is a q
• ng is a quick way to derive the and matrices from the graph that describes the m
• y, to use the graph to check if the matrices are correct. Background I took Udac
• a set of linear equations = X. The matrices and encode three type of constraint
• ved in the constraint. Once the and matrices are set up, we can solve for the ve
• matrix
• he value of the constraint while the matrix encodes the locations that are invol
• as a low variance of 1/W = 0.2). The matrix To build the matrix, we could start
• 1/W = 0.2). The matrix To build the matrix, we could start by finding its diago
• i and node j: Thus we have the Omega matrix: Also, an easy way to verify that th
• Also, an easy way to verify that the matrix is correct without drawing the graph
• matter
• ads and got 'Dave' and 'George' in a matter of a few hours. I also found a 3rd c
• matters
• h-Tzu, rounds up the family. Family matters We wanted our kids to learn as much
• measurement
• aph that describes the motion and measurement constraints or, alternatively, to
• in CS-373, the robot motions and measurements are written as a set of linear e
• the location of the robot and the measurement. Finally, depending on our confid
• itial position: x 0 = -3 Relative measurement: L = x 0 +10 Relative pose: x 1 =
• ative pose: x 1 = x 0 +5 Relative measurement: L = x 1 +5 Relative pose: x 2 =
• ative pose: x 2 = x 1 +3 Relative measurement: L = x 2 +1. Weight = 5 In this e
• . The robot moves twice, taking a measurement from every location: It starts at
• he first two locations. This last measurement, though, has a weight of 5 indica
• ng that it is a particularly good measurement (i.e., it has a low variance of 1
• igning a large weight to the last measurement. By assigning it a weight of 5 we
• ndicated that the variance of the measurement was 1/w = 0.2, i.e., a very good
• was 1/w = 0.2, i.e., a very good measurement, when in reality, it was the oppo
• ng a weight of 0.0001 to the last measurement, indicating that the measurement
• measurement, indicating that the measurement was lousy and the equation should
• measurements
• in CS-373, the robot motions and measurements are written as a set of linear eq
• media
• ds to problems because mail tags and media tags also use the "href" attribute b
• members
• le from JPL. I know four of its six members and think very highly of them: Mike
• message
• enough chads to be able to read the message of the page. This task that should
• s puzzle was What is the deciphered message? Not so easy, right? Give it a try.
• gether. Unlike puzzles 1 and 2, the message in this puzzle was a drawing, not a
• ere the people that had written the message of the puzzle. We did not know how
• ferences embeded in the text of the message. Anatole Litvak directed two movies
• woman I loved". The greeting in the message, "Comrade", was the clue that "Tova
• zle 2 4 pts: What is the deciphered message? I overthought this question too mu
• too much. Initially, because of the message, I thought that the cipher was a si
• simple one, the enciphering in the message was even simpler. There are 6 numbe
• 24, 18 and 21. At the bottom of the message, they show you what to do with the
• pond to "MEZ" Hence, the deciphered message is "JGOMEZ" Puzzle 3 2 pts: What is
• thought that the coordinates in the message, which I had not been able to put t
• method
• ut assembling it all, using a manual method. At this time, I was 5th in the lead
• mid-level
• alibration, image enhancement) and mid-level vision (e.g., object detection, re
• middle
• ds to recognize the structure in the middle as a pedestal. From then on it was s
• mind
• l was so different from what I had in mind: an automated tool based on constrain
• mistake
• le direction to solve puzzle 4. Big mistake! Although it was not possible to as
• month
• ed a challenge to put together, in a month, five shredded documents, each one p
• monty
• ion of words. As seen in the class, "monty" and "python" might be relevant but
• and "python" might be relevant but "monty python", as a combination, has a dif
• motion
• es from the graph that describes the motion and measurement constraints or, alte
• of SLAM taught in CS-373, the robot motions and measurements are written as a s
• here is an error associated with the motion and thus, we really do not know when
• 0 1 We can form a chain of relative motions of this kind, each related to the n
• motions
• of SLAM taught in CS-373, the robot motions and measurements are written as a s
• 0 1 We can form a chain of relative motions of this kind, each related to the n
• movies
• message. Anatole Litvak directed two movies in 1937: "Tovarich" and "The woman I
• mybook
• a book we might have: white_list = {"mybook", "chapter"} black_list = {"copyrigh



• name
• the puzzle. I participated using the name 'Goldsong'. This page describes my ex
• ible points. I participated using the name 'Goldsong' and managed to place 10th.
• ociated with the puzzle: What are the name and/or initials of the collaborators?
• ipedia. Puzzle 4 12 pts: What are the name and/or initials of the collaborators?
• ore for 'PK'. There were a total of 5 names. You would get 1, 3, 5, 8 or 12 if y
• 2 if you found 1, 2, 3, 4 or 5 of the names, respectively. Departing thoughts Th
• names
• re for 'PK'. There were a total of 5 names. You would get 1, 3, 5, 8 or 12 if y
• if you found 1, 2, 3, 4 or 5 of the names, respectively. Departing thoughts Th
• narrow
• e was a drawing, not a text. You can narrow posible characters that follow a par
• natural
• thing related to intelligence, both natural and artificial, e.g., brain organiz
• zation, theories of early learning, natural language processing, image understa
• nature
• or an online book, which by its very nature, it is bound to change everytime tha
• negative
• ments: for each edge i, j it is the negative of the weight between node i and no
• node
• and outgoing edges to the particular node: Then we can find the off-diagonal el
• is the negative of the weight between node i and node j: Thus we have the Omega
• tive of the weight between node i and node j: Thus we have the Omega matrix: Als
• each element of that corresponds to a node is the sum of the weighted incoming e
• of the weighted incoming edges to the node minus the sum of the weighted outgoin
• note
• document, a piece of data, a release note, etc. The situation boils to this: wh
• number
• crawling, a version that limits the number of pages visited and a breath-first
• ic. The function receives as input a number of files containing words that are v
• numbers
• ssage was even simpler. There are 6 numbers underlined in the text: 10, 23, 8,
• rn with the last three. The first 3 numbers correspond to the characters "JGO".



• object
• ncement) and mid-level vision (e.g., object detection, recognition and identific
• observation
• at allows a robot to use repeated observations of landmarks to locate himself i
• one by x i+1 = x i + d i i+1 The observation of the landmarks from different l
• close to 2+5=7, which matches the observation from the first location. Finally,
• o 5+1=6, which does not match the observations from the first two locations. Th
• instead of 7, like the first two observations indicated. Also, the robot did n
• it was the opposite. If the last observation had been weighted as bad instead
• er errors than before and all the observations are now consistent. Taking the c
• observations
• at allows a robot to use repeated observations of landmarks to locate himself in
• o 5+1=6, which does not match the observations from the first two locations. Thi
• instead of 7, like the first two observations indicated. Also, the robot did no
• er errors than before and all the observations are now consistent. Taking the ca
• observes
• n: It starts at x=-3, from which it observes the landmark at a distance of +10.
• en the robot moves +5 from which it observes the landmark at +5. This means that
• y, the robot moves +3 from which it observes the landmark at +1. This means that
• obtain
• up, we can solve for the vector X to obtain the absolute locations of the robot,
• ted entries as the starting point to obtain a final index. This step is importan
• obvious
• e that I like to work alone, it was obvious that it would have been great to ha
• off-diagonal
• icular node: Then we can find the off-diagonal elements: for each edge i, j it i
• offen
• und. Intranets and business websites offen attempt to emulate this facility pro
• d by a website and when it is, it is offen offered as a map of the website, whi
• offer
• ion that most word processors do not offer out-of-the-box. After taking CS-101,
• contents. This facility is not often offered by a website and when it is, it is
• website and when it is, it is offen offered as a map of the website, which is
• lready in 5th place and declined the offer. However, eventually they did catch
• omit
• tation. Also, the program is bound to omit words that the user considers importa
• he user definitely want to have or to omit respectively. For example, if the wor
• ones
• ssues instead of the long-term common ones. Others Some activities carried out d
• online
• y website and my essays about the an online book that I am writing. I submitted
• the contents of my website and of an online books that I am writting, a function
• nly two tools to create indexes from online documents: HTML indexer , a commerci
• ish task of creating an index for an online book, which by its very nature, it i
• th all the words that will be in the online index. create_html_index This final
• lected. The web page is ready to put online. By default, it is a strict XHTML 1.
• opportunity
• structure that I had not have the opportunity to use. Based on the contents of
• organization
• tural and artificial, e.g., brain organization, theories of early learning, natu
• orientation
• ictive ink color, hand writing or orientation that stood out. Unbeknownst to th
• with distinctive hand writing or orientation. Instead I started to work on the
• otherwise
• entify learning dissabilities that otherwise would not have been noticed until
• aving such a program to aid in the otherwise hellish task of creating an index
• out-of-the-box
• ost word processors do not offer out-of-the-box. After taking CS-101, I wrote an
• output
• of urls that need to be indexed. The output of this function is a type of index
• e, for the case of localization. The output of this function, available for insp
• ombinations of words on the fly. The output of this function is an list with all
• overthought
• What is the deciphered message? I overthought this question too much. Initially
• overview
• aper-bound books give a broad-topic overview with a table of contents. This faci



• package
• uments: HTML indexer , a commercial package that runs on Windows, and XRefHT ,
• page
• ove Dr. Pasadena, CA 91109 my JPL web page You can contact me at andres@acastano
• full-time dad of three toddlers. This page is pretty plain, all written manually
• ode for both Netscape IE. I had a web page up and running around the mid 90s, wh
• pated using the name 'Goldsong'. This page describes my experience with the cont
• iles via a GUI or, even better, a wep page but we did not learn how to set up su
• , a version that limits the number of pages visited and a breath-first version t
• breath-first version that limits the pages visited to those that fall within a
• e crawl_web routine are: Restrict the pages visited using white- and black-site
• l its chapters but will not crawl the pages devoted to the copyright or the glos
• osary. Memoization of all the visited pages to speed up the crawling and the fun
• the index. Deal with the retrieval of pages for which " a href=" in insuficient.
• nd "href" is prefered, there are many pages which still use uppercase tags and a
• tribute but they do not link to a web page. In the event that the pages that the
• to a web page. In the event that the pages that the user wants to index are dis
• index This final function makes a web page using the list of words that we selec
• st of words that we selected. The web page is ready to put online. By default, i
• By default, it is a strict XHTML 1.0 page in the english language that could pa
• h a good grade. The appereance of the page can be changed by hand, modifying the
• e changed by hand, modifying the html page or, without altering the page by modi
• he html page or, without altering the page by modifying a CSS file. This work is
• to be able to read the message of the page. This task that should not have taken
• discuss my results at the end of the page. Puzzle 2 The second puzzle had about
• pages
• a version that limits the number of pages visited and a breath-first version t
• breath-first version that limits the pages visited to those that fall within a
• crawl_web routine are: Restrict the pages visited using white- and black-site
• its chapters but will not crawl the pages devoted to the copyright or the glos
• sary. Memoization of all the visited pages to speed up the crawling and the fun
• he index. Deal with the retrieval of pages for which " a href=" in insuficient.
• d "href" is prefered, there are many pages which still use uppercase tags and a
• to a web page. In the event that the pages that the user wants to index are dis
• paper
• st one, it had been written in lined-paper, with two ink colors and had a coffe
• lor, not two, and was drawn on white paper, not lined one. However, the worst p
• paper-bound
• books-on-line. At the end of any paper-bound text book or reference book we fi
• parameter
• x The input to these functions are parameter files that can be modified by the
• user would have had access to the parameter files via a GUI or, even better, a
• parent
• and cannot be crawled from a single parent, the user can instead write the list
• park
• tand. The location was the Jose Marti park in Cienfuegos. A specific search for
• Cienfuegos. A specific search for the park showed that the dome was that of the
• drawing was based on the image of the park from Wikipedia. Puzzle 4 12 pts: What
• partial
• I had to have many simultaneous and partially redundant canvases. Still, I had
• ow posible characters that follow a partial word but this is not easy with a dr
• ). Can you figure them out from the partially-assembled puzzle? Puzzle 4 It did
• partially
• had to have many simultaneous and partially redundant canvases. Still, I had g
• . Can you figure them out from the partially-assembled puzzle? Puzzle 4 It did
• participation
• to nothing marking the end of my participation in the contest. I never got to g
• path
• ations of the robot, i.e., the actual path that it followed, and the location of
• pattern
• write the tool? Well.. there was a pattern that appeared useful. Puzzle 1 had
• with the first three. Continue the pattern with the last three. The first 3 nu
• pedestal
• ze the structure in the middle as a pedestal. From then on it was smooth sailing
• om then on it was smooth sailing. A pedestal implies a statue. An image search f
• people
• was interested in understanding how people solve complex puzzles. In 2011, they
• st in the forum about SLAM that many people found interesting. Hopefully, somebo
• eproduce a post about SLAM that many people found useful, in the hope that new C
• was interested in understanding how people solve complex puzzles. They issue th
• it had already started and when many people had already solved this puzzle. This
• umably, the "collaborators" were the people that had written the message of the
• team, the "Herded Rappers", made of people from JPL. I know four of its six mem
• personal
• edicted, we focus on our short-term personal issues instead of the long-term com
• pertinent
• set of words that are likely to be pertinent. In the tests that we ran, it is u
• pieces
• sing it so I spend less time pulling pieces into Keynote and more time pondering
• me it was ever more difficult to put pieces together. Unlike puzzles 1 and 2, th
• me about 100 hrs. to assemble enough pieces to answer the two questions of this
• utational tool that help them select pieces. You could ask the tool things like
• he tool things like "Show me all the pieces with the top 20% dark and the bottom
• place
• g the name 'Goldsong' and managed to place 10th. This is a summary of my experi
• iven that I did not have the tool in place, it was clear to me that I would not
• me to join them I was already in 5th place and declined the offer. However, eve
• atch up with me and finished in 13th place. In retrospective, it would have mad
• points
• ich scored at least 1 of 50 possible points. I participated using the name 'Gold
• at assembling the puzzle gave you no points. Only answering the question gave yo
• Only answering the question gave you points. Puzzle 3 Puzzles 1 and 2 definitely
• poor
• n hindsigh, my selection of tools was poor. Keynote is a very good presentation
• portion
• w do you measure solving the "easy" portion of the challenge? Find out answerin
• posible
• drawing, not a text. You can narrow posible characters that follow a partial wo
• position
• rement of a landmark The only known position of the robot is given by the Absolu
• t, commonly assigned to the initial position of the robot, i.e., x 0 , although
• eve it could be assigned to any one position. Suppose that the robot moves to a
• Our best guess, though, is that the position x 1 is about x 1 = x 0 + d 0 1 We c
• aints. A landmark L j observed from position x i at a distance k is described as
• the location set with the absolute position constraint. The setup Distances are
• ins shown on the graph are: Initial position: x 0 = -3 Relative measurement: L =
• column. In the case of the initial position, the sum should add up to the weigh
• add up to the weight of the initial position constraint. In the example, when ad
• post
• . Udacity - CS-373 - Autonomous car - post on SLAM I took Udacity 's CS-373 clas
• is finished. Here I am reproducing a post in the forum about SLAM that many peo
• very active forum. Here I reproduce a post about SLAM that many people found use
• prepares
• fun, gets you closer to your kids, prepares them for the future, and allows you
• prepositions
• ex, e.g., pronouns, conjunctions, prepositions, contractions, common verbs, etc.
• presentation
• was poor. Keynote is a very good presentation tool but it is no drawing program
• presently
• castano.com Hi.. I am an engineer, presently taking some years off while my chi
• pretty
• dad of three toddlers. This page is pretty plain, all written manually. It is m
• lot of hints. Hence, it looked also pretty easy to do by hand. The time will co
• uzzle 3 should take 40 hrs to solve, pretty doable. Thus, I had my answer and Bo
• previous
• ogressively more difficult than the previous one. There were 9000 entries, 69 of
• e was twice as hard to solve as the previous one. If that was the case, puzzle 3
• private
• ive forum that probably will become private once the course is finished. Here I
• probably
• access to a very active forum that probably will become private once the course
• problem
• cation of x 1 is also absolute. The problem is that there is an error associate
• vinding a local search engines. The problem is that unless you already know the
• earching for "href" itself leads to problems because mail tags and media tags a
• ot expecting such tool to solve the problem but I was expecting it to find an a
• problems
• earching for "href" itself leads to problems because mail tags and media tags al
• process
• writting, a function that most word processors do not offer out-of-the-box. Aft
• r of 1500-1000 words. The filtering process is completely automatic. The functi
• processors
• ritting, a function that most word processors do not offer out-of-the-box. After
• produce
• ieves the general concordance index produced by the web crawler to produce a sm
• ndex produced by the web crawler to produce a small set of words that are likel
• program
• the contents of the class I wrote a program that could index my website and my
• that I am writing. I submitted the program as an entry to the open contest tha
• king CS-101, I wrote an index maker program based on the search engine function
• indows and Java. I think that a new program can easily make an impact in the ar
• e me could benefit of having such a program to aid in the otherwise hellish tas
• This step is important because the program is not capable to measure the impor
• a different connotation. Also, the program is bound to omit words that the use
• sentation tool but it is no drawing program and thus, using it as a canvas to a
• progressively
• ive shredded documents, each one progressively more difficult. Points were awar
• ve shredded documents , each one progressively more difficult than the previous
• pronouns
• ly to not belong in an index, e.g., pronouns, conjunctions, prepositions, contra
• purple
• r is composed of functions, shown in purple in the block diagram: Crawl_web_by_k
• puzzles
• erstanding how people solve complex puzzles. In 2011, they issued a challenge t
• erstanding how people solve complex puzzles. They issue the Shredder Challenge
• a summary of my experience with the puzzles and my results. Puzzle 1 Like many
• han 8 hrs with a suitable tool. The puzzles only needed to be completed to the
• more tricky than the solving of the puzzles. For this puzzle, the question was
• useful chads. It appeared that the puzzles had been selected so that each one
• cult to put pieces together. Unlike puzzles 1 and 2, the message in this puzzle
• python
• . As seen in the class, "monty" and "python" might be relevant but "monty python
• python" might be relevant but "monty python", as a combination, has a different



• qualms
• start and a simple puzzle, I had no qualms doing it by hand, in the most unsoph
• query
• return a valid result to an arbitray query, i.e., "a", "as", "at", this it not
• questions
• lving the puzzle but for answering questions related to the puzzle. I participa
• challenge? Find out answering the questions in red Summary The Defense Advance
• found the answering of many of the questions more tricky than the solving of th
• r, I had no idea how to answer the questions associated with the puzzle. And ea
• le enough pieces to answer the two questions of this puzzle: What is the indica
• completely feasible to answer the questions without assembling it all, using a



• reality
• ., a very good measurement, when in reality, it was the opposite. If the last o
• realize
• nd a 3rd collaborator but I did not realize it, so I did not submit it as an an
• an you spot it? Likewise, I did not realize that all the answers stood out from
• could have gotten 19 pts. if I had realized that I had an additional answer in
• reasonable
• e suggested word list, it would be reasonable to black list "Albert" and white l
• others and hopefully settles on a reasonable configuration. However, I did not
• receives
• completely automatic. The function receives as input a number of files containi
• recognition
• l vision (e.g., object detection, recognition and identification). However, in
• recognize
• ually, I assembled enough chads to recognize the structure in the middle as a p
• reconfigurable
• In robotics, I have worked with reconfigurable robots and hand-to-eye coordinat
• reduction
• that we ran, it is usual to have a reduction in the size of the indexes by a fa
• redundant
• ve many simultaneous and partially redundant canvases. Still, I had gotten bett
• reference
• t d 0 1 . Since x 0 is an absolute reference, this would mean that the location
• nd of any paper-bound text book or reference book we find an index with words t
• n you are looking for a datum in a reference you do not look at the table of co
• references
• th "Litvak" and "1937" were needed references embeded in the text of the message
• relationship
• in which, again, we have that the relationship is only an approximation because
• relative
• x 0 + d 0 1 We can form a chain of relative motions of this kind, each related
• For every other constraint, either relative pose or landmark, the sum should be
• release
• ver, a document, a piece of data, a release note, etc. The situation boils to t
• relevant
• se entries were deemed likely to be relevant to a user looking for specific info
• lass, "monty" and "python" might be relevant but "monty python", as a combinatio
• remove
• to interact with the list and add or remove words or combinations of words on th
• reproduce
• ess to a very active forum. Here I reproduce a post about SLAM that many people
• requires
• o, injured with an in-utero stroke, requires a lot of stimulation. We find that
• respect
• rand Challenge and Urban Challenge, respectively. The student of this course ha
• location of the landmarks, all with respect to the location set with the absolu
• respectively
• d Challenge and Urban Challenge , respectively. The students of this course had
• nd Challenge and Urban Challenge, respectively. The student of this course had a
• efinitely want to have or to omit respectively. For example, if the words "Alber
• und 1, 2, 3, 4 or 5 of the names, respectively. Departing thoughts The contest w
• result
• ed but made significant errors. This result is a consequence of assigning a larg
• ht of W = 1/variance = 0.2, then the results would have been X = [-3.000, 2.050,
• , 5.000, 7.000] T i.e., the expected result for an error-free scenario. This wor
• e being indexed (shown in blue). The result of the four routines, shown in red,
• a search engine will return a valid result to an arbitray query, i.e., "a", "as
• results
• t of W = 1/variance = 0.2, then the results would have been X = [-3.000, 2.050,
• experience with the puzzles and my results. Puzzle 1 Like many other contestan
• er to the question? I'll discuss my results at the end of the page. Puzzle 2 Th
• retrieval
• ll create the index. Deal with the retrieval of pages for which " a href=" in i
• retrospective
• e and finished in 13th place. In retrospective, it would have made sense to joi
• return
• found. Although a search engine will return a valid result to an arbitray query,
• robot
• an Thrun, one of the most successful roboticists in many years, whose autonomou
• ing. It is a technique that allows a robot to use repeated observations of land
• ersion of SLAM taught in CS-373, the robot motions and measurements are written
• dmark The only known position of the robot is given by the Absolute Position co
• igned to the initial position of the robot, i.e., x 0 , although I believe it c
• o any one position. Suppose that the robot moves to a new location by an amount
• thus, we really do not know when the robot ended up at. Our best guess, though,
• e errors in both the location of the robot and the measurement. Finally, depend
• obtain the absolute locations of the robot, i.e., the actual path that it follo
• taken straight from the course, the robot is moving along a line, on which the
• the landmark L is also located. The robot moves twice, taking a measurement fr
• ndmark is close to -3+10=7. Then the robot moves +5 from which it observes the
• landmark at +5. This means that the robot should be close to -3+5= +2 and the
• rom the first location. Finally, the robot moves +3 from which it observes the
• landmark at +1. This means that the robot is at location 2+3 = 5 and the landm
• wo observations indicated. Also, the robot did not move +5 and +3 as commanded
• 0, 6.950] T Now everything fits. The robot moved as commanded with smaller erro
• roboticists
• Thrun, one of the most successful roboticists in many years, whose autonomous c
• Thrun, one of the most successful roboticists in many years, whose autonomous c
• robotics
• ible. Science Engineering I work in robotics and computer vision. In robotics, I
• in robotics and computer vision. In robotics, I have worked with reconfigurable
• robots
• s, I have worked with reconfigurable robots and hand-to-eye coordination, using
• the web. While I was working with my robots in the first floor of the Beckman In
• rounds
• , all adopted. Cosita, our Shih-Tzu, rounds up the family. Family matters We wan
• routine
• wn in blue). The result of the four routines, shown in red, is a HTML file read
• keyword The starting point for this routine is the crawl_web function from Udac
• hat we had to make to the crawl_web routine are: Restrict the pages visited usi
• routines
• wn in blue). The result of the four routines, shown in red, is a HTML file ready
• rule
• rly is extraordinarily sensible. As a rule, it is easier to teach anything to a
• runs
• L indexer , a commercial package that runs on Windows, and XRefHT , a freeware w



• satisfaction
• utomated tool based on constraint satisfaction. I was not expecting such tool to
• scenario
• e expected result for an error-free scenario. This work is licensed under a Crea
• score
• There were 9000 entries, 69 of which scored at least 1 of 50 possible points. I
• ranked is 7th, very distant from the score of 50 pts. earned by the winning tea
• script
• dacity Search Engine Index Maker - a script to create indexes of intra-nets and
• Index Maker Index Maker is a Python script that semi-automates the creation of
• classes. The following is a brief description of the four main functions. crawl
• search
• der Challenge The Defense Advanced Research Project Agency - DARPA - was interes
• use the topic of understanding how a search engine worked sounded to alluring to
• an index maker program based on the search engine functions seen in the course.
• ate this facility provinding a local search engines. The problem is that unless
• the website you do not know what to search for. Paper-bound books give a broad-
• ercase tags and attributes. Finally, searching for "href" itself leads to proble
• very single string found. Although a search engine will return a valid result to
• n red Summary The Defense Advanced Research Project Agency - DARPA - was interes
• pedestal implies a statue. An image search for statues in Cuba gave me one in w
• Marti park in Cienfuegos. A specific search for the park showed that the dome wa
• second
• at the end of the page. Puzzle 2 The second puzzle had about 350 chads but unlik
• seen
• based on the search engine functions seen in the course. Download code Examples
• mportance of combination of words. As seen in the class, "monty" and "python" mi
• seldom
• ed as a map of the website, which is seldom used. Instead, the tool of choice of
• select
• sion of Power Point. In hindsigh, my selection of tools was poor. Keynote is a v
• t appeared that the puzzles had been selected so that each one was twice as hard
• a computational tool that help them select pieces. You could ask the tool thing
• select_index_entries
• keyword Suggest_index_entries select_index_entries Create_html_index The input t
• ing for specific information. select_index_entries This function uses the index
• selection
• on of Power Point. In hindsigh, my selection of tools was poor. Keynote is a ve
• semi-automates
• ex Maker is a Python script that semi-automates the creation of indexes for intr
• sense
• In retrospective, it would have made sense to join forces but even in that case
• sensible
• the blank) early is extraordinarily sensible. As a rule, it is easier to teach a
• ed puzzle? Puzzle 4 It did not look sensible to attack puzzle 4 by hand. Too man
• settles
• trains all the others and hopefully settles on a reasonable configuration. Howe
• setup
• he absolute position constraint. The setup Distances are shown in blue, weights
• ion of the equations is correct, the setup of the equations had an error. Let's
• short-term
• d to be predicted, we focus on our short-term personal issues instead of the lon
• sieves
• suggest_index_entries This function sieves the general concordance index produc
• significant
• e +5 and +3 as commanded but made significant errors. This result is a conseque
• similar
• rom different locations x i lead to similar constraints. A landmark L j observe
• simpler
• enciphering in the message was even simpler. There are 6 numbers underlined in
• simultaneous
• mited in size, I had to have many simultaneous and partially redundant canvases.
• single
• jointed and cannot be crawled from a single parent, the user can instead write t
• ic type of index that contains every single string found. Although a search engi
• situation
• of data, a release note, etc. The situation boils to this: when you are lookin
• slashdot
• nts, I found about the challenge in slashdot , after it had already started and
• smooth
• e as a pedestal. From then on it was smooth sailing. A pedestal implies a statue
• society
• l issues I am an acid critic of our society. We have it all backwards. Even now
• solution
• weight to an equation, leading to a solution in which the error for that particu
• ould count more towards finding the solution of the system. Likewise, an equatio
• hted outgoing edges: Comment on the solution This example was given in class but
• expecting it to find an approximate solution that could be useful. Thus, it coul
• somebody
• eople found interesting. Hopefully, somebody else will find it useful too.
• specific
• wn as a "concordance", i.e., a very specific type of index that contains every s
• o be relevant to a user looking for specific information. select_index_entries T
• he Jose Marti park in Cienfuegos. A specific search for the park showed that the
• specify
• ite and black lists of entries that specify words or combination of words that
• spend
• ill, it is not unusual for a user to spend a large amount of time traversing th
• I had gotten better at using it so I spend less time pulling pieces into Keynot
• spite
• stitution Caesar cipher. However, in spite that the Caesar cipher is a very sim
• d have really benefited me. Also, in spite that I like to work alone, it was ob
• spot
• nd did not get credit for it. Can you spot it? Likewise, I did not realize that
• y/town? The country is easy enough to spot: Cuba. However, the city/town is very
• stain
• this puzzle: assemble the coffee mug stain, assemble the borders, assemble the
• stands
• y students also find it useful. SLAM stands for Simultaneous Localization and Ma
• starts
• measurement from every location: It starts at x=-3, from which it observes the
• statue
• smooth sailing. A pedestal implies a statue. An image search for statues in Cuba
• mplies a statue. An image search for statues in Cuba gave me one in which you co
• statues
• plies a statue. An image search for statues in Cuba gave me one in which you co
• stimulation
• n-utero stroke, requires a lot of stimulation. We find that teaching our kids t
• stores
• find the location where the website stores what s/he is looking for: a driver,
• straight
• 1. Weight = 5 In this example taken straight from the course, the robot is movin
• strict
• make to the crawl_web routine are: Restrict the pages visited using white- and b
• y to put online. By default, it is a strict XHTML 1.0 page in the english langua
• stroke
• Bella who, injured with an in-utero stroke, requires a lot of stimulation. We f
• structure
• to learn about Hash tables, a data structure that I had not have the opportunit
• ser to its destination is its tree structure, which often is sufficient; still,
• bled enough chads to recognize the structure in the middle as a pedestal. From
• student
• Urban Challenge, respectively. The student of this course had access to a very
• in the hope that new CS-373 Udacity students also find it useful. SLAM stands f
• students
• Urban Challenge , respectively. The students of this course had access to a very
• in the hope that new CS-373 Udacity students also find it useful. SLAM stands fo
• submit
• t I did not realize it, so I did not submit it as an answer and did not get cred
• subset
• ble for inspection as a file, is the subset of the concordance whose entries wer
• substitution
• ught that the cipher was a simple substitution Caesar cipher. However, in spite
• successful
• y Sebastian Thrun, one of the most successful roboticists in many years, whose a
• y Sebastian Thrun, one of the most successful roboticists in many years, whose a
• sufficient
• its tree structure, which often is sufficient; still, it is not unusual for a us
• suggest_index_entries
• expect to find in an index. suggest_index_entries This function sieves the gen
• suitable
• t have taken more than 8 hrs with a suitable tool. The puzzles only needed to be
• summary
• nd managed to place 10th. This is a summary of my experience with the puzzles a
• surprise
• to alluring to me. The class was a surprise: fast paced, interesting and I even
• surroundings
• ironment built as it explores its surroundings. In Graph-SLAM, the version of SL
• swim
• nd that teaching our kids to read, to swim, or to _____ (you fill out the blank)
• system
• towards finding the solution of the system. Likewise, an equation associated wi
• d be given no value when solving the system, we would have gotten: X = [-3.000,



• table
• s give a broad-topic overview with a table of contents. This facility is not of
• n a reference you do not look at the table of contents; you look at the index.
• at the index. The information in the table of contents is too broad. Instead, t
• tables
• g and I even got to learn about Hash tables, a data structure that I had not hav
• tags
• many pages which still use uppercase tags and attributes. Finally, searching fo
• itself leads to problems because mail tags and media tags also use the "href" at
• problems because mail tags and media tags also use the "href" attribute but the
• taken
• = x 2 +1. Weight = 5 In this example taken straight from the course, the robot
• page. This task that should not have taken more than 8 hrs with a suitable tool
• target
• impact in the area of indexing. The target user for this application would be o
• task
• ogram to aid in the otherwise hellish task of creating an index for an online bo
• to read the message of the page. This task that should not have taken more than
• rly it was not the best tool for this task: I could not zoom in and out of the c
• taught
• c Car - in March 2012. The class was taught by Sebastian Thrun, one of the most
• c Car - in March 2012. The class was taught by Sebastian Thrun, one of the most
• . In Graph-SLAM, the version of SLAM taught in CS-373, the robot motions and mea
• teach
• s a lot of stimulation. We find that teaching our kids to read, to swim, or to
• sensible. As a rule, it is easier to teach anything to a 2-year old than a 5-ye
• ow Python (which CS-101 was going to teach) and because the topic of understand
• team
• core of 50 pts. earned by the winning team. Puzzle 1 2 pts: What is the appropri
• ave been great to have been part of a team. During the contest I was in touch wi
• e contest I was in touch with another team, the "Herded Rappers", made of people
• winning and many of the high ranking teams, had put together a computational to
• teams
• winning and many of the high ranking teams, had put together a computational to
• technique
• Localization and Mapping. It is a technique that allows a robot to use repeate
• test
• ndex of CS-101 forum entry for the contest What is Index Maker Index Maker is a
• at are likely to be pertinent. In the tests that we ran, it is usual to have a r
• nguage that could pass any validation test with a good grade. The appereance of
• tests
• t are likely to be pertinent. In the tests that we ran, it is usual to have a r
• text
• n-line. At the end of any paper-bound text book or reference book we find an ind
• e in this puzzle was a drawing, not a text. You can narrow posible characters th
• were needed references embeded in the text of the message. Anatole Litvak direct
• There are 6 numbers underlined in the text: 10, 23, 8, 24, 18 and 21. At the bot
• theories
• tificial, e.g., brain organization, theories of early learning, natural language
• thoughts
• the names, respectively. Departing thoughts The contest was a lot of fun and ve
• tiff
• ay possible: I cut the chads from the tiff images in which they came and pasted
• time
• s carried out during the copious free time left from being a full-time dad of th
• ious free time left from being a full-time dad of three toddlers. This page is p
• for a user to spend a large amount of time traversing the tree, trying to find t
• ry nature, it is bound to change everytime that a change is made. Block diagram
• d also pretty easy to do by hand. The time will come when I would need to put to
• en better at using it so I spend less time pulling pieces into Keynote and more
• pulling pieces into Keynote and more time pondering how to connect them. There
• associated with the puzzle. And each time it was ever more difficult to put pie
• er, the worst part is that for a long time the drawing was not clear. In the end
• t all, using a manual method. At this time, I was 5th in the leaderboard. Unfort
• t have been able to solve puzzle 5 on time. Results I got a total of 17 pts. I c
• y difficult to figure out. For a long time I thought that the coordinates in the
• title
• question was What is the appropriate title being referenced? Well.. I am giving
• zle 1 2 pts: What is the appropriate title being referenced? You needed to solv
• toddlers
• from being a full-time dad of three toddlers. This page is pretty plain, all wri
• tool
• e, which is seldom used. Instead, the tool of choice of a website to guide a use
• t we want to have. There are only two tools to create indexes from online docume
• r Point. In hindsigh, my selection of tools was poor. Keynote is a very good pre
• . Keynote is a very good presentation tool but it is no drawing program and thus
• taken more than 8 hrs with a suitable tool. The puzzles only needed to be comple
• d need to put together a computatinal tool to help me assemble the puzzle but th
• Keynote. Clearly it was not the best tool for this task: I could not zoom in an
• it together. Indeed, a computational tool was not needed for this puzzle either
• efinitely did no need a computational tool to solve them. Puzzles 4 and 5 defini
• olve it by hand or start to write the tool? Well.. there was a pattern that appe
• started to work on the computational tool that seemed the only viable direction
• le to assemble the puzzle without the tool, it was completely feasible to answer
• owever, given that I did not have the tool in place, it was clear to me that I w
• ams, had put together a computational tool that help them select pieces. You cou
• them select pieces. You could ask the tool things like "Show me all the pieces w
• as it was, I was surprised that their tool was so different from what I had in m
• from what I had in mind: an automated tool based on constraint satisfaction. I w
• atisfaction. I was not expecting such tool to solve the problem but I was expect
• uration. However, I did not finish my tool and words are wind.
• tools
• we want to have. There are only two tools to create indexes from online docume
• Point. In hindsigh, my selection of tools was poor. Keynote is a very good pre
• topic
• was going to teach) and because the topic of understanding how a search engine
• that tell where a particular word or topic is found. Intranets and business web
• for. Paper-bound books give a broad-topic overview with a table of contents. T
• total
• ve puzzle 5 on time. Results I got a total of 17 pts. I could have gotten 19 pt
• gotten 2 more for 'PK'. There were a total of 5 names. You would get 1, 3, 5, 8
• touch
• a team. During the contest I was in touch with another team, the "Herded Rappe
• tree
• uide a user to its destination is its tree structure, which often is sufficient;
• a large amount of time traversing the tree, trying to find the location where th
• tribute
• ain, all written manually. It is my tribute to the early days of the web. While
• tricky
• wering of many of the questions more tricky than the solving of the puzzles. For
• twice
• k L is also located. The robot moves twice, taking a measurement from every loc
• d been selected so that each one was twice as hard to solve as the previous one
• twins
• three children: Sofi, now 4 1/2, and twins Danny and Bella, age 3, all adopted.
• type
• ns = X. The matrices and encode three type of constraints: Absolute Position Rel
• xed. The output of this function is a type of index commonly known as a "concord
• "concordance", i.e., a very specific type of index that contains every single s
• .e., "a", "as", "at", this it not the type of information that we expect to find



• unknown
• e going to need it. Puzzle 3 was an unknown. Should I solve it by hand or start
• unusual
• ten is sufficient; still, it is not unusual for a user to spend a large amount
• uppercase
• ere are many pages which still use uppercase tags and attributes. Finally, sear
• urls
• he user can instead write the list of urls that need to be indexed. The output o
• user
• ool of choice of a website to guide a user to its destination is its tree struct
• cient; still, it is not unusual for a user to spend a large amount of time trave
• t in the area of indexing. The target user for this application would be one tha
• ter files that can be modified by the user (shown in green) and the html files o
• uploaded to the website. Ideally, the user would have had access to the paramete
• In the event that the pages that the user wants to index are disjointed and can
• be crawled from a single parent, the user can instead write the list of urls th
• contractions, common verbs, etc. The user can opt to use all, some or none of t
• ere deemed likely to be relevant to a user looking for specific information. sel
• ogram is bound to omit words that the user considers important and to add words
• s important and to add words that the user does not care for. The function allow
• not care for. The function allows the user to alter the suggested entries in two
• ested entries in two ways. First, the user can set white and black lists of entr
• ords or combination of words that the user definitely want to have or to omit re
• appens automatically, after which the user is allow to interact with the list an



• valid
• though a search engine will return a valid result to an arbitray query, i.e., "
• english language that could pass any validation test with a good grade. The app
• validation
• glish language that could pass any validation test with a good grade. The appere
• value
• ight (0 W 1). The vector encodes the value of the constraint while the matrix e
• get 0 - 1 + 6 - 5 = 0 The vector The value of each element of that corresponds
• and the equation should be given no value when solving the system, we would ha
• variance
• lar equation has a small error (low variance) and should count more towards find
• associated with a large error (high variance) should be assigned a low weight (0
• ood measurement (i.e., it has a low variance of 1/W = 0.2). The matrix To build
• a weight of 5 we indicated that the variance of the measurement was 1/w = 0.2, i
• tead of good, i.e., if it had had a variance of 5 and a weight of W = 1/variance
• variance of 5 and a weight of W = 1/variance = 0.2, then the results would have
• vector
• e assigned a low weight (0 W 1). The vector encodes the value of the constraint
• ces are set up, we can solve for the vector X to obtain the absolute locations o
• x 2 in we get 0 - 1 + 6 - 5 = 0 The vector The value of each element of that co
• verbs
• , prepositions, contractions, common verbs, etc. The user can opt to use all, s
• verify
• e Omega matrix: Also, an easy way to verify that the matrix is correct without d
• version
• ts surroundings. In Graph-SLAM, the version of SLAM taught in CS-373, the robot
• ndows, and XRefHT , a freeware with versions for both Windows and Java. I think
• unction from Udacious. We saw three versions of it: indiscriminate crawling, a
• s of it: indiscriminate crawling, a version that limits the number of pages vis
• of pages visited and a breath-first version that limits the pages visited to th
• crawl. Unfortunatelly, none of this versions is useful to travel exclusively a
• d pasted them in Keynote, the Apple version of Power Point. In hindsigh, my sel
• versions
• ndows, and XRefHT , a freeware with versions for both Windows and Java. I think
• unction from Udacious. We saw three versions of it: indiscriminate crawling, a v
• crawl. Unfortunatelly, none of this versions is useful to travel exclusively a g
• viable
• putational tool that seemed the only viable direction to solve puzzle 4. Big mis
• vision
• ring I work in robotics and computer vision. In robotics, I have worked with rec
• coordination, using manipulators. In vision, I work in low-level vision (e.g., i
• tors. In vision, I work in low-level vision (e.g., image formation, camera calib
• on, image enhancement) and mid-level vision (e.g., object detection, recognition



• warm-up
• had about 200 chads, i.e., it was a warm-up. With a late start and a simple puz
• website
• wrote a program that could index my website and my essays about the an online b
• eate an index of the contents of my website and of an online books that I am wr
• wnload code Examples: index of this website index of CS-101 forum entry for the
• ic is found. Intranets and business websites offen attempt to emulate this faci
• ou already know the contents of the website you do not know what to search for.
• facility is not often offered by a website and when it is, it is offen offered
• it is offen offered as a map of the website, which is seldom used. Instead, the
• d. Instead, the tool of choice of a website to guide a user to its destination
• ying to find the location where the website stores what s/he is looking for: a
• o to create a simple index of their website but still sees the benefits of havi
• ML file ready to be uploaded to the website. Ideally, the user would have had a
• websites
• ic is found. Intranets and business websites offen attempt to emulate this facil
• weigh
• n a particular constraint we can add weight to an equation, leading to a soluti
• *L j = W*x i + W*k We assign a large weigh to an equation (i.e., W 1) when we b
• h variance) should be assigned a low weight (0 W 1). The vector encodes the val
• e setup Distances are shown in blue, weights are shown in red. The constrains s
• This last measurement, though, has a weight of 5 indicating that it is a partic
• nt of the diagonal is the sum of the weights of the incoming and outgoing edges
• edge i, j it is the negative of the weight between node i and node j: Thus we
• sition, the sum should add up to the weight of the initial position constraint.
• responds to a node is the sum of the weighted incoming edges to the node minus
• ges to the node minus the sum of the weighted outgoing edges: Comment on the so
• s a consequence of assigning a large weight to the last measurement. By assigni
• last measurement. By assigning it a weight of 5 we indicated that the variance
• te. If the last observation had been weighted as bad instead of good, i.e., if
• if it had had a variance of 5 and a weight of W = 1/variance = 0.2, then the r
• case to the extreme and assigning a weight of 0.0001 to the last measurement,
• weight
• n a particular constraint we can add weight to an equation, leading to a solutio
• h variance) should be assigned a low weight (0 W 1). The vector encodes the valu
• e setup Distances are shown in blue, weights are shown in red. The constrains sh
• This last measurement, though, has a weight of 5 indicating that it is a particu
• nt of the diagonal is the sum of the weights of the incoming and outgoing edges
• edge i, j it is the negative of the weight between node i and node j: Thus we h
• sition, the sum should add up to the weight of the initial position constraint.
• responds to a node is the sum of the weighted incoming edges to the node minus t
• ges to the node minus the sum of the weighted outgoing edges: Comment on the sol
• s a consequence of assigning a large weight to the last measurement. By assignin
• last measurement. By assigning it a weight of 5 we indicated that the variance
• te. If the last observation had been weighted as bad instead of good, i.e., if i
• if it had had a variance of 5 and a weight of W = 1/variance = 0.2, then the re
• case to the extreme and assigning a weight of 0.0001 to the last measurement, i
• weights
• setup Distances are shown in blue, weights are shown in red. The constrains sh
• t of the diagonal is the sum of the weights of the incoming and outgoing edges
• white
• re: Restrict the pages visited using white- and black-site lists. For example,
• mple, to index a book we might have: white_list = {"mybook", "chapter"} black_l
• in two ways. First, the user can set white and black lists of entries that spec
• easonable to black list "Albert" and white list the combination "Albert Einstei
• tself, would not. The merging of the white and black lists with the suggested w
• ink color, not two, and was drawn on white paper, not lined one. However, the w
• the top 20% dark and the bottom 40% white". This looked very useful. Useful as
• white-
• re: Restrict the pages visited using white- and black-site lists. For example, t
• white_list
• le, to index a book we might have: white_list = {"mybook", "chapter"} black_list
• wife
• ars off while my children grow up. My wife Becky and I have three children: Sofi
• wind
• ad available to put in the challenge dwindle to nothing marking the end of my pa
• did not finish my tool and words are wind.
• woman
• movies in 1937: "Tovarich" and "The woman I loved". The greeting in the messag
• word
• t I am writting, a function that most word processors do not offer out-of-the-bo
• reference book we find an index with words that tell where a particular word or
• th words that tell where a particular word or topic is found. Intranets and busi
• in the block diagram: Crawl_web_by_keyword Suggest_index_entries select_index_en
• four main functions. crawl_web_by_keyword The starting point for this routine i
• web crawler to produce a small set of words that are likely to be pertinent. In
• o-3, i.e, if the concordance had 3000 words, the index of suggested words is lik
• ad 3000 words, the index of suggested words is likely to have on the order of 15
• ely to have on the order of 1500-1000 words. The filtering process is completely
• as input a number of files containing words that are very likely to not belong i
• sure the importance of combination of words. As seen in the class, "monty" and "
• n. Also, the program is bound to omit words that the user considers important an
• e user considers important and to add words that the user does not care for. The
• d black lists of entries that specify words or combination of words that the use
• that specify words or combination of words that the user definitely want to hav
• mit respectively. For example, if the words "Albert" and "Einstein" are in the s
• " and "Einstein" are in the suggested word list, it would be reasonable to black
• te and black lists with the suggested words happens automatically, after which t
• eract with the list and add or remove words or combinations of words on the fly.
• dd or remove words or combinations of words on the fly. The output of this funct
• this function is an list with all the words that will be in the online index. cr
• on makes a web page using the list of words that we selected. The web page is re
• ible characters that follow a partial word but this is not easy with a drawing.
• Hence, I did not try to assemble the words with distinctive hand writing or ori
• However, I did not finish my tool and words are wind.
• words
• reference book we find an index with words that tell where a particular word or
• eb crawler to produce a small set of words that are likely to be pertinent. In
• -3, i.e, if the concordance had 3000 words, the index of suggested words is lik
• d 3000 words, the index of suggested words is likely to have on the order of 15
• ly to have on the order of 1500-1000 words. The filtering process is completely
• s input a number of files containing words that are very likely to not belong i
• ure the importance of combination of words. As seen in the class, "monty" and "
• . Also, the program is bound to omit words that the user considers important an
• user considers important and to add words that the user does not care for. The
• black lists of entries that specify words or combination of words that the use
• that specify words or combination of words that the user definitely want to hav
• it respectively. For example, if the words "Albert" and "Einstein" are in the s
• e and black lists with the suggested words happens automatically, after which t
• ract with the list and add or remove words or combinations of words on the fly.
• d or remove words or combinations of words on the fly. The output of this funct
• his function is an list with all the words that will be in the online index. cr
• n makes a web page using the list of words that we selected. The web page is re
• Hence, I did not try to assemble the words with distinctive hand writing or ori
• owever, I did not finish my tool and words are wind.
• work
• ossible as early as possible. This is working well with all of them but in parti
• onger possible. Science Engineering I work in robotics and computer vision. In r
• computer vision. In robotics, I have worked with reconfigurable robots and hand
• ion, using manipulators. In vision, I work in low-level vision (e.g., image form
• he early days of the web. While I was working with my robots in the first floor
• sult for an error-free scenario. This work is licensed under a Creative Commons
• he page by modifying a CSS file. This work is licensed under a Creative Commons
• part is that all of these approaches worked equally well. In the end, it took m
• or orientation. Instead I started to work on the computational tool that seemed
• ted me. Also, in spite that I like to work alone, it was obvious that it would h
• worst
• e paper, not lined one. However, the worst part is that for a long time the dra
• wrong
• Thus, I had my answer and Boy, was I wrong! At the beginning it looked as if I



• years
• m an engineer, presently taking some years off while my children grow up. My wi
• most successful roboticists in many years, whose autonomous cars placed 1st an
• most successful roboticists in many years, whose autonomous cars placed 1st an



• zoom
• best tool for this task: I could not zoom in and out of the canvas and, since t