MAS-Soccer: A testbed based on multi-agent cooperation
Chunyi Shi · Journal of Tsinghua University(Science and Technology) · 2005
A testbed is needed to test and compare various multi-agent cooperative problem solving algorithms. A multi-agent cooperative simulation system based on BDI reasoning, MAS-Soccer, was developed to emphasize the importance of joint strategies in multi-agent systems, rather than the performance details of specific actions. The system can be used to test reinforcement learning algorithms based on feature vector extraction and the traditional Q-learning algorithm in the Robocup free kick game. Tests illustrate the efficiency of the system and that MAS-Soccer can efficiently and correctly compare various cooperative strategies.