Eco-Be robots playing a game of mixed reality Pac-Man

Undergraduate students at the University of Manitoba have developed a mixed reality demo of CITIZEN's centimeter high Eco-Be robots playing Pac-Man. The Eco-Be is a series of tiny robots no taller than one inch, and they are powered using watch batteries. As you can see from the photo below, these wheeled robots are really small and obviously only capable of navigating smooth surfaces (click the image to enlarge.)


The Eco-Be robots are controlled by a remote computer using infrared communication. Researchers at the autonomous agents laboratory, University of Manitoba, have been developing a mixed reality platform for educational and robotics use. It is currently an important component of the computer science department's undergraduate curriculum. As a result, several students developed a mixed reality application in which the Eco-Be robots play a game of Pac-Man. You can watch the demonstration in the following video (although be warned that the robots were having difficulties due to drained batteries.)

Evac-Op: distributed decision making using constraint optimization

Work at Drexel University focuses on the important problem of disaster evacuation support for large numbers of people in poor communication conditions. If anything, the recent disaster at New Orleans makes the development of such technologies a necessity. The Drexel team has developed the Evac-Op software as a testbed for distributed decision making in such terrible situations.

Evac-Op is a prototype system for assisting emergency personnel in monitoring and conducting evacuation and sheltering operations. It is a novel application of distributed constraint optimization combined with mobile wireless networking to sharing situation information and making global decisions on issues such as shelter assignments. In addition to exploring this application and new applications of distributed constraint optimization, Evac-Op is intended as a vehicle for investigating distributed decision making under poor communications, uncertainty, and change.

The team has been busy demonstrating a current prototype that utilizes tablet PCs or PDAs communicating over a Wi-Fi network for coordinating the evacuation of a large city in an intelligent way. Evac-Op compiles information, such as the availability of medical supplies, from several shelters scattered around the city and the needs of evacuees; it then directs the evacuees to the best shelter where they can be taken care off avoiding overcrowding and wasting medical supplies.

You can find the slides from the AAAI demonstration of Evac-Op here (pdf.)

Speech recognition + robot = hilarity

The How to say "No" to a robot video is easily the funniest entry in the first AI video competition that was held earlier this week during the AAAI 2007 conference on artificial intelligence. In an earlier post, I described my experiences with speech recognition software both commercial and open source. Hendrik Zender, Patric Jensfelt, and Oscar Martinez Mozos from various institutes in Germany and Sweden have put together a demonstration video that shows the state-of-the-art of speech recognition systems in robotics; their results match my own experiences very accurately (and I am sure that other robotics researchers would say the same.) If you have ever tried to use the speech recognition function of SONY's AIBO robot pet then you will be able to guess how the video plays out. The punchline is just plain hilarious!

Semantic Robot Vision Challenge video and results

The first Semantic Robot Vision Challenge (SRVC) held during the AAAI 2007 conference on artificial intelligence has now concluded. A total of 3 teams participated in the robot league and 2 teams in the software league. The SRVC is a new research competition that encourages the integration of high-level perceptual systems and robots.

The competition was a form of robotic scavenger hunt with a twist. First, each team was given a textual list of objects unknown to them until the competition's start. Then, within a 30-minute time limit, the teams were allowed to use any online service such as Google or Flickr to construct appearance models of the objects. At the conclusion of this stage, each team's robot was allowed 15 minutes in a home-like environment to autonomously search and photograph as many of the specified objects as possible. After all robots finished, the teams had another 30 minutes to segment and label the objects within the collected images. Lastly, the judges received the labeled images and evaluated the performance of each robot.

The software-only competition involved the same steps except that the judges used a reference robot to manually collect images for the teams.

Four university teams participated in the challenge. The teams were from the University of British Columbia, Kansas State University, University of Maryland and Princeton.

The following video shows the robot exploration and data gathering phase for the 3 robots that participated in the robot competition. The final standing are given at the end of this post, after the video.

Semantic Robot Vision Challenge Final Standings

Robot League

1. University of British Columbia


2. University of Maryland


3. Kansas State University

Software League

1. Princeton


2. Kansas State University

Congratulations to all the teams. The SRVC was a great success and the organizers hope to continue the challenge in the years to come.

Man wins the first man-machine poker challenge

After two intense days of poker and 2000 hands played, the human team consisting of Ali Eslami and Phil Laak are the winners of the first ever man-machine poker championship. The artificial intelligence software Polaris put up a great fight but it couldn't win in the final and most important session. Specifically, the championship was played over a period of 2 days and 4 sessions (2 each day.) The 1st session ended in a draw while Polaris came out the winner at the end of the 2nd session. The human team recovered on the 3rd session with a win so that the final result would be determined at the 4th session played tonight. Ali and Phil topped Polaris by a combined $570 in the final session to win the man-machine championship. The event was educational more than anything else and I feel pleasure that I was able to watch it live at the AAAI conference. The Alberta team will probably be very happy with Polaris' performance and I am sure they now have lots of data to analyze in their efforts to create the perfect poker playing machine. Until next time!

Our previous coverage of the poker challenge event can be found here.

Additional information can also be found at the official challenge website here.

DARwIn: Dynamic Anthropomorphic Robot with Intelligence

As part of the AAAI-07 conference robot exhibition and demonstration, Karl Muecke from the Robotics Mechanisms Laboratory (RoMeLa) at Virginia Tech, demonstrates their humanoid robot DARwIn. First, he plays a game of role the dice with the robot and then he shakes hands with it. The robot is instructed via text commands written on paper (as you can see in the video.) Finally, DARwIn showcases his soccer skills by kicking a ball with grace and style; the robot participated in the recently held Robocup and it was the first and only US entry in the humanoid division. The Virginia Tech team hopes to eventually reduce the cost and time of building these robots so that they can mass produce them and sell them in the consumer market.

Machine versus man poker challenge is on!

Probably the most exciting event during the 2nd day of the AAAI conference on artificial intelligence was the start of the first man-machine poker championship. The artificial intelligence program Polaris from the University of Alberta will be playing two days of poker against professional poker players Ali Eslami and Phil Laak. Ali and Phil will be playing against two instances of Polaris in different rooms. The game is Texas Hold'em and it will be played in a way that luck will be a lesser factor to the final outcome. Instead of explaining this in writing, watch the following video of Polaris creator Jonathan Schaeffer introducing the players and explaining the importance of the challenge in addition to how the games will be played.

The twenty-second AAAI conference on Artificial Intelligence (AAAI-07)

Next week, I will be attending the 22nd AAAI conference on artificial intelligence held in Vancouver, Canada; other than the fact that the weather in Vancouver is terrible (it is raining at the moment,) attending the conference should be lots of fun. I'll try to summarize some of the most interesting events on this blog. So, come back during the week to learn about the many exciting happenings at AAAI-07. Here is a summary of some of the events,

Technical program: There is a huge number of technical papers that will be presented during the conference; it looks like the conference has 8 simultaneous tracks.

Invited speakers: I am really looking forward to the invited talks. Other than Alan Mackworth's presidential address on Agents, Bodies, Constraints and Dynamics, there will be talks by Lise Getoor on Graph Identification; Geoffrey S.F. Ling on Revolutionizing Prostheses; Michaels Wooldridge on Logic for Automated Mechanism Design; Oren Etzioni on AI in a Moore's Law World; Alan C. Schultz on Moving Toward Peer-to-Peer Human-Robot Interaction; Toby Walsh on Representing and Reasoning about Preferences and Matt Brown on Big A, Small I: Smart Ends from Simple Means (it is about AI in computer games.)

Exhibition: A number of book publishers and companies that depend on AI will have booths at the conference. Publishers include the AAAI Press, Cambridge University Press, Springer, MIT Press, and Morgan and Claypool Publishers. Companies include Google, Microsoft and TextDigger.

AI and Robot competitions: There are many competitions taking place during the conference. These include the new AI video competition, general game playing competition, trading agent competition, semantic robot vision challenge and, of course, the man versus machine poker challenge. I will post photos and, hopefully, some video from these events. I am really looking forward to the poker challenge, video challenge and the robot competition and exhibition. AAAI will also host a number of intelligent systems demonstrations.

Awards: As usual, AAAI will be presenting a number of awards to outstanding research and fellows. Some of the awards include the Robert S. Engelmore Memorial Award and Lecture, Classic Paper Award, Distinguished Service Award, and the IJCAI-JAIR Best Paper Prize.

It is going to be a very exciting week!

Chinook: the unbeatable checkers playing AI

It's all over the news since yesterday. Jonathan Schaeffer's game group at the University of Alberta, Canada, has announced that they have finally solved checkers. Solving the game required 18 years of 50-200 computers crunching numbers; Schaeffer started the computation in 1989. Don't be surprised about the length of time this took because the game's search complexity is an astonishing 500,995,484,682,338,672,639 positions (that's 500 billion billion!)

So, what is the solution? According to the official and newly lunched Chinook website,

From the standard starting position, Black (who moves first) is guaranteed a draw with perfect play. White (moving second) is also guaranteed a draw, regardless of what Black plays as the opening move.

If you think that you can defeat Chinook, then head over to the project website and play a game (or more) against it. The software seems to allow only 50 simultaneous games so you might have a hard time scheduling one; I couldn't but I'll be back to try it next week when all the traffic from the recent media exposure subsides.

So, what is next for Schaeffer and his team? They have a man versus machine poker challenge scheduled for next week during the annual AAAI conference. Their poker playing software Polaris will be playing against professional poker players Phil Laak and Ali Esmali. I'll be attending the conference and I will post information about the poker challenge and all the other robot competitions. It should be lots of fun.

Related articles:
Computers Solve Checkers—It's a Draw
Alberta professor crowns ultimate checkers king
Computers crack famous board game

Multi-agent systems and game theory

I often find that those of us who study Computer Science fail to realize that much of the tools that we use have been developed in other disciplines and only recently adapted to our field. One example is the use of Markov Decision Processes (MDPs) to model sequential decision making problems; MDPs were initially developed in Operations Research well before we begun to use them in AI. Another example is the study of multi-agent systems using Game Theory. Game Theory was developed for use in economics and applied mathematics. Professors do mention the origin of these theories when they teach their courses but students often tend to ignore them.

At any rate, this post was mostly inspired because of a documentary on evolutionary biology that I found on YouTube. The documentary called "Nice Guys Finish First" features Richard Dawkins explaining how cooperation among agents is prevalent in nature. Dawkins, of course, is the well known evolutionary biologist and most outspoken member of the modern atheist movement. The documentary is from 1987 and it nicely explains the use of Game Theory to describe how multi-agent systems work in nature. If you have taken a multi-agent systems course then you are probably familiar with the prisoner's dilemma (PD) problem that exposes the dynamics of a game played between two agents faced with the choice of cooperating or not; cooperation leads to higher reward but rational agents would choose not to cooperate.

In the documentary, Dawkins explains how species play this game repeatedly; in this case, the Tit for Tat strategy is the best to play. It is a strategy that favors cooperation and punishes defection. If you want to know more about the different ways of playing the game and the best strategies for each then click here (Stanford Encyclopedia of Philosophy.)

The documentary is worth watching but make sure you have plenty of free time because it is 45 minutes long.

i-LIMB Hand is a breakthrough prosthetic hand for amputees

Scientists at Touch Bionics in Scotland have designed, tested and brought to market a new state-of-the-art bionic hand that promises an incredible amount of precision and control for patients.

According to the company's press release,

Touch Bionics, developer of the world’s first commercially available bionic hand, today announced that its i-LIMB Hand and ProDigits partial hand prostheses are now generally available and have been successfully fitted to a significant number of patients across the United States and in Europe.

Touch Bionics’ i-LIMB Hand looks and acts like a real human hand and is the world’s first widely available prosthetic device with five individually powered digits. In another industry first, Touch Bionics’ ProDigits product is adapted for patients who have a partial hand, due either to congenitally missing fingers or fingers lost through an accident. Partial hand is an area of prosthetics that has been without suitable powered products in the past.

The prosthetic hand is controlled via the electric signals generated by the muscles in the patient's remaining, healthy portion of the arm.

The i-LIMB Hand offers a unique, highly intuitive control system that uses a traditional myoelectric signal input to open and close the hand’s life-like fingers. Myoelectric controls utilize the electrical signal generated by muscles in the remaining portion of a patient’s limb. This signal is picked up by electrodes that sit on the surface of the skin. Users of existing, basic myoelectric prosthetic hands are able to quickly adapt to the system and can master the device’s new functionality within minutes.

Another important aspect of the device is that the fingers automatically adjust their positions depending on the object grasped (this is demonstrated in the second of the videos linked to below.) In addition, other than the fact that this new bionic hand is easy to use, it is also easy to repair. According to the information posted on the Touch Bionics website, a patient can swap any broken finger on the hand by removing a single screw cutting down on repair time that would normally require shipping the prosthetic back to the manufacturer.

You can watch videos of the bionic hand in use here (interview with the inventor of the bionic hand,) here (interviews with several patients who use the hand and Touch Bionics personnel,) and finally here (lots of demonstrations of the the i-LIMB Hand's capabilities.)

Photos are copyright Touch Bionics.

Get the scoop on robotics from CBC

The Canadian Broadcasting Corporation (CBC) is running, on their website, an In Depth technology special with a focus on robotics. In a series of articles, CBC writers are exploring the past, present and future of robotics. Following is a summary of some of the most interesting articles.

They start with an attempt to define what a robot is. The article titled "What is a robot?" talks about the origin of the word "robot" and looks up its definition in the Oxford English Dictionary. They call upon a number of prominent North American roboticists to give a definition for robot; these include Joseph Engelberger who is known as the father of robotics; Alan Mackworth who is current president of AAAI and he is known as the founding father of robot soccer; the well known Rodney Brooks from MIT and lastly professor Gregory Dudek from McGill. Obviously, they all have a slightly different definition of what exactly constitutes a robot so it makes the article worth reading. CBC invites readers to try and give their own definition of a robot here.

Tara Kimura's article "Domestic Helpers?" explores how and when robots will become useful general purpose machines that help around the house. Several experts are called upon to talk about failed and successful projects in the consumer robot market. Kimura gives a good list of consumer robots currently available for sale including Robosapien, Roomba, Scooba, Pleo, and the Lego Mindstorms NXT.

A third article titled "Warning! Robots ahead" discusses the issues around the future of robots and their role in society as they become commonplace and begin to offer valuables services such as caring for the elderly. UBC's Richard Rosenberg discusses the ethical implications of robotics as robots become ubiquitous in our daily lives. This is a topic that has received much attention in the last couple of years as many governments are putting together committees to prepare guidelines for the safe and proper usage of robots in society.

Lastly, CBC writer Martin Morrow explores the fascination of pop culture with robots in his article "Dream Machines." Morrow surveys some of the most well known, loved and/or hated robots in books, movies, television, music, and cartoons. I am sure that many of this blog's readers are familiar with most if not all the robots mentioned in this article.

When you are done reading these articles, test your knowledge of robots with CBC's Robot quiz. See if you can best my score 9 out of 10; I made a mistake on the last question. Don't cheat by looking up the answers on Google. Post your results here if you like.

Thanks goes to Robert Sim for alerting us about the CBC articles.

Smart Machines is one year old

Well, happy birthday to us!

Our tiny blog today turns one year old. It was on July 16th, 2006, when I made the very first post announcing the birth of the Smart Machines blog with a focus on Artificial Intelligence and Robotics news. The first real post was the next day about CMU receiving 15 million dollars in funding. The very first comment was posted on August 1st, 2006, on a story about the new Lego Mindstorms NXT robot kit going on sale.

Since those days, the blog has grown considerably. We are now read by more than 600 people daily. Nearly half of the readers receive the newest post via email or RSS while the rest prefer to visit the blog. We have experienced a steady 20% growth month after month for the last 6 months. It was a little rocky at the beginning before we begun to rank well in search engines; most notably, Google sends us a considerable chunk of traffic every day.

Last but not least, I want to thank all of our readers. Thank you for reading our blog and commenting on the articles. We always welcome your comments and criticism if you have any. When I started this blog, I was hoping to build a nice community of experts and amateurs who would help each other learn about AI and robotics motivated by the latest news in the field. The community is still small but continuously growing which gives me the strength to continue posting. I can't really thank you enough for this!

Happy Birthday!

Runbot: the fastest walking robot

At least 3 people emailed me yesterday to alert me about BBC's article on Runbot. The article describes Tao Geng's bipedal dynamic walking robot. Robots that are well known such as Honda's ASIMO operate using the Zero Moment Point criterion for stability control and motion generation which is a passive walking technique difficult to scale up to fast moving robots. The dynamic walking approach addresses this issue mimicking nature.

The Runbot stands 23cm tall and has four actuated joints using RC servo-motors. Sensors on Runbot's feet are used to detect contact with the ground. The robot's design is relatively low cost and simple. Most notably, Runbot uses reinforcement learning techniques to improve on its walking abilities and adapt its gait to different terrains.

One of the major problems with such dynamic biped walkers is that even though it is possible to achieve good balance and speed once moving, making the transition from standing to walking and vice verse is still very challenging. Runbot is also just a planar walker so it can only move on a straight line which is a much simpler problem to solve. Also, Runbot and other similar robots under development, are not currently capable of high-level trajectory planning and obstacle avoidance which are, of course, necessary skills for an autonomous robot. But researchers are making great progress in developing dynamic bipeds and it won't be long before the above issues are also addressed.

In the meantime, you can watch videos of Runbot in action at Tao Geng's website here.

Robots at Play 2007 festival

The 2007 edition of the Robots at Play international festival is just over one month away. The festival is held annually in Odense, Denmark, and it is a 2-day event covering all aspects of robotics including lectures from Computer Scientists, robot competitions, exhibition, film and robot art.

In fact, Robots at Play rewards a 10,000 Euro prize for the "promotion of a deeper understanding and use of robotics in everyday life since robotic systems are playing an increasingly important role in people’s lives all over the world." Last year's winner was the teddy robot Huggable and developed at MIT.

The festival also includes a number of Lego Mindstorms competitions for young kids with an interest in robotics. The arts exhibition "brings to the light a number of meaningful artistic applications in robotics by looking at both pure installations, interactive installations, performances and interactive performances."

Of interest is also the Robo[Trash] competition.

Assisted by competent professionals from Odense Technical College, and your own technical abilities, you may construct a robot the size of a human being from old trash, a washing machine and advanced micro computers. Your "trash robot" will compete with other candidates in terms of technical capability, appearance etc. Therefore, your robot should move around in a funny and different way and have a unique appearance, if it is to do well in the Robo [Trash] competition.

The Robots at Play festival sounds like it is great fun for scientists, kids and just about anyone who is curious about how robotics is making its way into our daily lives.

The festival takes place August 23-25th.

Robot soccer: Then and now!

Just a couple of days ago, we had the conclusion of the 2007 edition of the annual RoboCup tournament held at Georgia Tech. The event which has exploded in popularity in recent years includes a variety of different soccer playing robots both physical and simulation as well as other events such as the robot rescue competition and numerous industrial and academic demonstrations.

You can find lots of video footage and photos from RoboCup events on YouTube and Flickr but I want to use this post to show you the progress that we have made in robotics during the last 15 years. Alan Mackworth from the University of British Columbia was the first person to suggest that soccer was the perfect domain for testing artificial intelligence. He talked about this in a couple of papers back in 1992 and 1993; because of that, the International RoboCup Foundation has named him "The Founding Father" of robot soccer. His research team at the Laboratory for Computation Intelligence at UBC constructed the first ever soccer playing robots, the Dynamo and Dynamites. So, without further delay, here is a video circa 1993 that showcases these robots using constraint programming to perform basic soccer maneuvers for goalkeeping and shooting.

So, how advanced is robot soccer today? 15 years later, small humanoid robots have become increasingly prevalent in RoboCup. From having simple demonstrations of humanoid robots kicking the ball just a 3 years ago, the event now hosts a number of 2 vs 2 humanoid events added to the 4-legged AIBO and the many wheeled robot leagues. As you can see from the 2007 RoboCup video bellow, we have come a long way since 1992.



The next 5-10 years are going to be very exciting. Hopefully, we will start to see more cooperation between the robots executing plays that require coordination and cooperation such as passing the ball or defending as a team. For that to happen, we will need better reasoning algorithms but also robot bodies that are more responsive and with better sensing. Considering the fast development of such technologies, I am fairly certain that such advanced soccer playing robots will soon become reality.

Internet search engines and artificial intelligence

A couple of days ago, I finished reading John Battelle's The Search: How Google and Its Rivals Rewrote the Rules of Business and Transformed Our Culture. The book is about the rise and fall of online search engines, a domain that for the last few years has been dominated by Google.

The book makes for some good reading and I highly recommend it to anyone who is interested in knowing how, in less than 10 years, companies such as Google and Yahoo became household names and an integral part of our lives.

But the reason I mention the book here is because Battelle mentions Artificial Intelligence and its relation to search in his book. In fact, he does so in the very first chapter in a section titled “Search as Artificial Intelligence?” As you may have guessed from the section's title, he is a bit off when it comes to understanding the link between AI and search.

Battelle mostly focuses on Google's efforts to create a natural language interface for their index,

"I would like to see the search engines become like the computers in Star Trek," Google employee number one, Craig Silverstein, quips. "You talk to them and they understand what you're asking."

Other search engines, most notably Ask Jeeves, have also pursued this goal and continue to do so with little success. Natural language understanding is an important field within AI but search is far more integral to AI than this.

The major components of an AI system are a knowledge base (KB,) an Inference/Reasoning Engine and a language that we can use to communicate with the computer. The KB (basically a database) makes it possible for us to write down knowledge/facts such as “the sky is blue” or “cars have 3 or 4 wheels.” The inference engine allows us to derive new facts from those in the knowledge base, or in other words, it gives us the ability to compute answers to questions posed in the system's language. Semantics is also a very important component but I won't go into it now because it is usually hard to explain in a couple of sentences and I want to talk about search.

So, how does search fit into an AI system?

Well, search is the engine underneath the hood that makes the inference/reasoning engine work.

For any interesting domain, the size of the KB and the additional facts that can be inferred from it comprise a huge space. Efficient search algorithms are important for navigating this space in order to answer users' queries. Algorithms such as Depth First Search (DFS,) Breadth First Search (BFS,) Heuristic Search and Stochastic Search are what make inference possible. Brute force methods such as DFS and BFS don't really work for large problems and this is why stochastic and heuristic search algorithms are the focus of most current research. For those of you have an undergraduate degree in Computer Science, I hope that you now know why we insist on teaching you these search algorithms. They are extremely important!

So, how does Google fit into all this?

Google is on a mission to index all of the data that is available on the World Wide Web, our desktop computer and also in our printed books and press. Not forgetting of course satellite and street level images and video. I believe their corporate motto is “Don't be evil” but their mission is to “organize the world's information and make it universally accessible and useful.” Google's index is in fact a huge KB and the well known PageRank (PR) value assigned to every page they index is a heuristic (one of many that Google uses) that makes it possible to efficiently search through this enormous KB. The language that we use to access the KB and the inference engine in order get our questions answered is the keyword-based interface on Google's simplistic homepage.

In other words, Google is and it has always been an AI system. The natural language interface that Battelle talks about in his book would be a mostly welcomed upgrade to the system's language, i.e., a more intuitive way for asking questions and probably for presenting answers back to us.

Ancient Greek programmable robot replica

The editors at New Scientist have constructed a replica of what is believed to be the earliest known programmable robot.

In about 60 AD, a Greek engineer called Hero constructed a three-wheeled cart that could carry a group of automata to the front of a stage where they would perform for an audience. Power came from a falling weight that pulled on string wrapped round the cart's drive axle, and Sharkey reckons this string-based control mechanism is exactly equivalent to a modern programming language.

By the way, Noel Sharkey is a computer scientist at the University of Sheffield, UK, who recently discovered that one of Leonardo da Vinci's robotic creations was based on Hero's designs.

Here is the video of the replica as constructed by the magazine's editors,

This robot does not really qualify as a real robot in my opinion because it cannot sense its environment and react to it. Regardless, considering that this automaton was designed and built for the first time almost 2,000 years ago, I have to say that I am really amazed by it. The Greeks had a fascination with autonomous mechanisms; one can find descriptions of robotic creatures in their mythology and philosophy. Greek engineers are also well known for having constructed a number of automata such as Archytas' mechanical bird that was propelled by steam. Also don't forget about the Antikythera Mechanism that was recently revealed to be an ancient analogue computer used for astronomical calculations.

New Nao at Robocup 2007

The 2007 edition of the world famous Robocup tournament is well under way and more than half way through at Georgia Tech in Atalanta; the competition ends on July 10th.

French humanoid robot maker, Aldebaran Robotics, was there to show off their second generation Nao robot. The new robot is a bit smaller than the first prototype and it boasts stereo vision. The company demonstrated the robot walking and kicking a small soccer ball. I heard rumors that after scoring a goal, Nao went on the hunt for an Italian to headbutt paying homage to soccer superstar Zinade Zidane (that was a poor attempt at a joke, btw.)

You can watch videos of the new Nao at the Aldebaran website here; the videos include the robot kicking the ball and a Haka dance (in simulation) which is common in rugby but not in soccer.

Other than soccer, Robocup also hosts a robot rescue competition and events for high school students interested in robotics. There is a Flickr group with lots of photos from the events in addition to a YouTube channel with a large number of videos.

Alan Mackworth from the University of British Columbia was the first to suggest that the game of soccer was a good testbed for AI algorithms. Playing soccer requires intelligence at the single agent level but also the cooperation of multiple agents in a very dynamic environment. Mackworth is currently the president elect of AAAI.

Meet RUBI: a robot designed for understanding social interaction

The Machine Perception Laboratory at the University of California, San Diego, is working on a very interesting robotics project developing a robot suitable for studies in social interaction. The robot is named RUBI and it consists of a wheeled robot base with a humanoid upper body. The robot is designed for interaction with young children; the research team behind RUBI hopes to use the robot in order to learn models of social interaction and behavior. Such models are necessary for the design and development of helper robots that one day will live among us.

RUBI stands 3 feet tall and has an expressive head with many degrees of freedom; on-board and off-board computers provide RUBI with all the necessary computational power for performing complex image and audio processing. The robot's upper body is complete with two arms which the team admits were the most difficult to design because of safety considerations when interacting with young children. RUBI's perceptual capabilities include face detection, tracking and some basic facial expression recognition. Kids can interact with RUBI via a touch screen interface.

So far, RUBI has undergone field testing interacting with 18 to 24 month old toddlers as part of their daily activities at the Early Childhood Education Center at the University of California, San Diego. You can read the details behind the history of the project and the team's progress at the 2-year mark in the recently published paper titled "The RUBI Project: A Progress Report. (pdf document)"

Photo of RUBI with kid outdoors is copyright the RUBI Project.

Having fun with speech recognition software

Last week, I mentioned that Sphinx was the best speech recognition software that I had used in my robotics projects. Not that it is perfect; far from it! It works better than other software that I have tested but it still requires that one speaks slowly and clearly using a good microphone.

Today, I came across a hilarious blog post written by Stephen Potter who is a program manager with Microsoft Windows. His post, titled "How to punish a speech recognition system" is a hilarious look at the state-of-the-art in speech recognition and how to mess with it. If you are trying to trick your computer then consider using one or more of the seven methods Stephen outlines in his post; some are easier to execute and other require a bit of practice.

Take for example tip number 5,

Pretend you're different people as the session progresses. Bit subtle this one, but in order to improve accuracy, speech recognizers like to decide early on what kind of speaker you are - male/female, child/adult, etc., and assume that you won't change. Nice try, reco-bot. This futile assumption can be wiped on the floor simply by first pretending to be a middle-aged man and then suddenly a twelve-year old girl! (You might want to practice voices beforehand.) A fun variant of this is to get different kinds of people together, and hand the phone between them at each dialog turn - great party game.

Read the complete list of tips on how to punish your speech recognition system on Stephen's Working the Spoken Word blog.

Artificial Intelligence and the game of Go

The online edition of the UK Times has published a very interesting article titled "Why computers can’t surpass Go and collect $1 million." Ben Macintyre is the article's author. He correctly points out that the ancient Chinese game of Go has far too many states, i.e., board configurations, that a brute force approach for finding the optimal strategy is a lost cause. He attributes the ability of humans to play the game well to our intellect that is characterized "by adaptation to uncertainty, intuition, wisdom, the ability to understand the thoughts and feelings of others, and a sense of mortality."

It is not clear to me what feelings and mortality have to do with playing Go and I don't really know how to quantify wisdom so I will not duel on these concepts. However, his claim that machines lack the ability to function under uncertainty, show intuition and extract patterns is only partly correct.

Artificial intelligence systems today perform well in a large number of very specialized tasks due to the use of probabilistic methods in modeling uncertainty in their actions and observations. In addition, the field of pattern recognition is quickly advancing and in many areas it's performance is vastly superior to human abilities (Google's search engine for example) but to be honest it is seriously lacking in others such as, for example, in object class recognition. In terms of intuition, I would say that for machines it essentially translates to heuristic functions that can guide the search for a solution.

Having the proper heuristic that prunes much of the game's search space is exactly what enables humans to play the game better than a machine today. In fact, it is exactly the reason why Deep Blue also managed to defeat Kasparov in the game of chess; it was not only a matter of brute force calculation as Macintyre claims in his article. Don't forget that just because a human can win in the game of Go (or any other game for that matter,) it doesn't mean that he is playing a perfect strategy; he simply plays better than his opponent. There is no reason to believe, that with the proper heuristics, a machine could not also play the game well enough to beat a junior Go champion and win the $1 million dollar challenge mentioned in the article.

Finally, I totally disagree with his closing statement that "Only when the machines surpass Go (and collect $1 million) will artificial intelligence truly be worthy to compete with the human kind." I think people used to make similar claims in the days before machines bested us in chess. Then again, finding the right heuristic to win in the game of Go doesn't necessarily mean that we have found the right heuristic for solving all problems. Unless we believe the claim that the game of Go holds the answer to the inner-workings of human intelligence.

PS: I couldn't find any information about the $1 million Go challenge but one of the commentators mentions that this might be the Ing Challenge which is no longer offered.