October 2003

RoboCup News

The RoboCup-2003 competitions were held in Padua Italy in July. There were a number of levels of competition, from simulation, with no real robots, up to humanoid. Humanoid robots just compete at penalty kicks and the like at this point.

Each year the rules of the competitions are upgraded to reflect the improved capabilities of the robot competitors, so I suspect the rules for humanoid robots will gradually evolve to the stated RoboCup goal of fielding a robotic team that can beat a human one.

The smaller robots are playing real soccer. The links below include some video clips that show them in action.

One of the smaller leagues is called the Small-Sized League. It's robots are coffee-can size and play on a surface about the size of ping pong table, using an orange golf ball. The robots are autonomous but radio controlled from computers, and global vision is shared among them, provided by cameras mounted above the playing field.

This year Cornell's Big Red team won, defeating the University of Queensland Australia's RoboRoos by a score of 1-0 in overtime. Given that Cornell had won a few years previous with a score of 10-0 and then 7-6, before losing a year, it appears that defenses are getting much better.

Cornell provides some of the best documentation I've found on robotic soccer, and sure enough, the design goals for the 2003 team included improved defense. They had special goalie programming introduced this year, and had modifications to the dribbling mechanisms that made it easier for their robots to strip the ball from opposing robots.

Cornell assembles a team of students and faculty at the beginning of each academic year, with the goal of fielding a better team in the Summer's competition. The team starts, like any good athletic team, by studying last year's game films.

Cornell's project goals provide a good insight into the state of the art. Some of the things they wanted to improve in 2003 were:

  • setup times,
  • maintenance of down robots,
  • onboard sensing to prevent missed passes,
  • better defense through strategy,
  • more accurate kicking and dribbling,
  • easier manufacturability,
  • adaptive game learning,
  • use of a goalie,
  • keep sources of funding going.

Note the emphasis on defense in this year's designm, explaining the low scoring results.

The Cornell project was divided into modular components. For example there was a group working on mechanical design, and that group was subdivided into a group working on the drive mechanism, the dribbling subsystem, and the kicking subsystem.

This means there were people working all year on just coming up with a better kicking mechanism for a coffed-can sized robot. They did, last years kicking arm warped on off center kicks, costing them accuracy.

The dribbling mechanism of Cornell's robots was accredited with being a major factor in previous year's victories. This year they were working on better control, pass reception, and the ability to steal the ball from opponents.

This was all done within a budget of $30,000 from corporate and alumni sponsors.

AI and the Law

It is tempting to look at applying AI to the law, using computer programs to determine points of law and thus avoiding some of the costs of human decision making.

Law is represented as rules, which seems to imply a rule-based technology; and as cases, which seems to imply a case-based reasoning technology. And reasoning is used deduce conclusions from the rules and cases, which seems to imply some sort of logic or theorem proving technology.

Research and experimental systems have been built using each of these, and the discussion on whether or not they are practical always revolves around the issues of fuzziness in the law. Michael Aikenhead's paper has some interesting thoughts on rule-base and case-based implementations of AI legal systems, and suggests they are limited.

As a point of criticism of rule-based systems, he uses this example of a law: no vehicles allowed in the park.

The problem, he notes, is what's a vehicle? Is a bicycle allowed? or a police car? While he argues the problem is difficult, it seems that this is just the sort of problem that an ontology could solve.

Looking through the titles of papers in the International Association for AI and Law, it appears that is exactly one of the hot areas of research. Mark Lauritsen's article goes into details on a law ontology, refining concepts such as "concept maps", which seem exactly what is needed to answer the question of what is a vehicle.

Michael Aikenhead also goes on to discuss the potential for argumentative systems. Robert Kowalski shows how to use logic programming for a theorem proving approach to computerizing law.


http://robocup.mae.cornell.edu/documentation/robocup/2003/2003ME.pdf - Home page for the Cornell RoboCup project.

http://robocup.mae.cornell.edu/documentation/robocup/2003/2003ME.pdf - The best documentation I've found so far on building a RoboCup robot. It's a fascinating story of engineering goals

http://www.icsi.berkeley.edu/~behnke/papers/FU-Fighters2002.pdf - A good overview document on the FU Fighters RoboCup team from Freie University, Berlin Germany.

http://www.java-video.de/stock99/ - Java applet video of the FU Fighters small robots in action.

http://arti.vub.ac.be/RoboCup/rules/rules.html - The rules for the small robot competitions.

http://www.iaail.org/ - International Association for AI and the Law home page. This is an organization you can join with a journal devoted to AI and the law.

http://www.denniskennedy.com/ailaw.htm - Dennis Kennedy's web site which is intended to be a starting point for locating resources on AI and the Law. Lots of good links.

http://qsilver.queensu.ca/law/smith.htm - An excellent article by J.C. Smith and Daphne Gelbart introducing the concepts of different types of legal reasoning and how they might be modelled by AI systems.

http://www.law.warwick.ac.uk/ltj/5-1c.html - Some interesting thoughts on whether AI can be applied to the law or not, by Michael Aikenhead of Collingwood College. Maybe some of his criticisms are based on rigid sorts of views of AI systems.

http://www.kluweronline.com/issn/0924-8463 - Home page for the Artificial Intelligence and Law journal. Not cheap.

http://www.bileta.ac.uk/ailaw/ailaw.html - The AI and Law Group of the British and Irish Legal Education Technology Association home page, with a number of good links.

http://www.capstonepractice.com/OntoOpen.html - Marc Lauritsen's article proposing ontologies for the law, with concepts like "concept maps".

http://elj.warwick.ac.uk/jilt/artifint/2hunter/contents.htm - And how about neural networks for encoding the law?

http://www-lp.doc.ic.ac.uk/UserPages/staff/rak/papers/law.html - Robert Kowalski's thoughts on how logic programming can be applied to the law.