The Universities of York and Leeds present

 

Tenth

Knowledge Representation and Reasoning

Distinguished Lecturer

 

Bernhard Nebel

 

 

Professor  Nebel's  talk  will  be  delivered  from Freiburg  to  the Universities  of  York and  Leeds simultaneously     by     video-conferencing    and access-grid  facilities.   The talk  is aimed at  a general  audience of  computer  scientists and  the public is invited to attend at either venue.

 

 

How to Win the World Cup

Action Selection for Robotic Soccer

 

13:30, Wednesday 15 June 2005

 

Haycock Lecture Theater (Room CS/103)

Department of Computer Science

University of York

                       

and

 

Informatics Conference Room (6.08)

with additional seating in the L8 Boardroom (8.01)

EC Stoner Building, School of Computing

University of Leeds

 

 

While soccer might  not be considered as the  best domain for applying Artificial Intelligence methods, it turns  out to be a challenging and appealing domain.  In this talk I  will focus on the  aspect of action selection techniques that  we have applied in our  robotic soccer team (which won  the Robot World Cup  Soccer Games three times)  and in our autonomous football table.

 

We employed a technique called behavior networks for action selection in our  robotic soccer  team CS Freiburg.   While this is  a technique that works  very well  in practice, there  are not many  known results about  their  theoretical properties.   In  particular,  there are  no guarantees that  a goal will be  reached if it is  reachable.  We were able to show  that under some conditions such  guarantees can be given and  that  the  topologies  of  the  networks  we  used  satisfy  this condition.  As a next step in trying to play soccer against humans, we developed an  autonomous football table,  where human and  machine can compete  against each  other  directly.  Here  we  used two  different action  selection strategies, namely  a simple  reactive scheme  and a decision-theoretic method.  It turns out that although the predictions made  by  the  decision  theoretic  scheme are  quite  crude,  it  has nevertheless an advantage over  the purely reactive method. Currently, we are  working on using reinforcement learning  techniques to advance the performance of the robot.

 

About the speaker: Bernhard Nebel is a Professor at the Albert-Ludwigs University in Freiburg, where he is  head of the research group on the foundations of  Artificial Intelligence. His  prolific research output has made major  contributions to the field of  symbolic AI, especially in  the areas  of spatial  and temporal  knowledge  representation and computational complexity.  Further details  of his many activities and publications can be found at www.informatik.uni-freiburg.de/~nebel/.

 

Professor Nebel  has also achieved huge  success as manager  of the CS Freiburg  Robotic Soccer  team,  three times  world  champions in  the international RoboCup competition.  See www.cs-freiburg.de for further information.

 

Recently he  and his students  have applied AI reasoning  and Computer Vision techniques to develop  a robotic table football opponent, which can beat 85%  of human players. This is now  available as a commercial product.

 

About the lecture series:  This  lecture series  is  sponsored  and organised by the  Department of Computer Science at  the University of York  and the School  of Computing  at the  University of  Leeds.  Its purpose  is  to  promote  the  strong  research  interests  that  both departments have  in knowledge representation  and reasoning.  Further information can be found at www.cs.york.ac.uk/aig/seminars/dist.html.

 

For further information contact either:


Alan Frisch

Department of Computer Science

University of York

York YO10 5DD

phone: +44 (1904) 432745

email: frisch@cs.york.ac.uk

 

Brandon Bennett

School of Computing

University of Leeds

Leeds LS2 9JT

phone: +44 (113) 343 1070

email: brandoncomp.leeds.ac.uk