Ulm-sparrows 2 Improved Robot Hardware

In RoboCup-98, sparrows team worked hard just to get both a simulation and a middle size robot team to work and to successfully participate in a major tournament. For this year, we were in a better position to start some more serious research work. Aside of improvements in the robot hardware and an extension of the vision processing capabilities, we implemented a more complete version of our soccer agent architecture and made some progress in the areas player localization, environment modelling, and basic playing skills. For the latter, we started to apply learning techniques. ULM-Sparrows is a research eeort seeking to investigate and solve open problems relevant to both the RoboCup Challenge KAK + 97] and a local interdisciplinary research eeort called SMART PK97] 1. Some research issues of particular interest to our team include skill learning in continuous domains, adaptive spatial modeling of highly dynamic environments, and emergent multiagent cooperation for achieving coordinated team play without explicit communication. We also have a general interest in studying robot control architectures for soccer agents and neurosymbolic integration, in particular the integration of symbolic and neural methods in robot control architectures. See KESo99] for a more detailed account of our goals. We have both a simulation team and a middle size real robot team to pursue these goals. In RoboCup-98, we used three modiied Pioneer-1 robots, a LEGO-based robot, and a custom-built goalie based on a toy tracked vehicle. The per-1 See www.uni-ulm.de/SMART/ for more information.