The Dynamo Project - robotic soccer

Alan Mackworth leads the Dynamite Testbed project, which is part of a larger robotics effort, the Dynamo Project at the University of British Columbia. The project's soccer robots, "Dynamites", are commercial radio-controlled toy cars which play on a table-sized field. An overhead camera views the action and feeds information to the robots' control programs.

Papers co-authored by Michael Sahota are available on his publications page. He tells me that they'll send their simulator software to those who request it.

Dynamite: A testbed for multiple mobile robots (1993, 7 pages)
Rod Barman, Stewart Kingdon, Alan K. Mackworth, Dinesh Pai, Michael Sahota, Heath Wilkinson, Ying Zhang
Postscript. An overview of the Dynamite hardware and software, and the purposes it's being put to.

Reactive deliberation: An architecture for real-time intelligent control in dynamic environments (1994, 7 pages)
Michael Sahota
Postscript. A reactive deliberation robot controller consists of two asynchronous components, an executor which interacts with the world in real time and a deliberator which plans ahead, selecting goals or abstract actions for the executor. Experiments with soccer robots suggest that this architecture's ability to change plans in mid-execution is valuable.

Can situated robots play soccer? (1994, 6 pages)
Michael Sahota and Alan K. Mackworth
Postscript. Argues that robotic soccer is a good common domain on which to focus the efforts of AI researchers. It's difficult enough to drive research but limited enough to offer achievable goals.

Real-time intelligent behavior in dynamic environments: Soccer-playing robots (1994, over 85 pages)
Michael Sahota
Compressed postscript. A master's thesis.

Reinforcement Learning
Craig Boutilier, Roger Ford and Keiji Kanazawa
This web page describes unpublished work showing that a reinforcement learning method converges quickly and produces good results when it's given a few special soccer features and some small rewards as "coaching tips".