Agents That Learn from Other Competitive Agents

Abstract

We have been doing a research on visionbased reinforcement learning and applied the method to build real soccer playing robots. In the rst stage [Asada et al., 1994a], a robot learned to shoot a ball into a goal. In the second stage [ Asada et al., 1994b; Asada et al., 1994c ] , we set up an opponent just before the goal, that is, a goal keeper, and make the robot learn to shoot a ball into a goal avoiding the goal keeper. This can be considered as a problem of learning from other agents. The big dierence from the existing schemes such as learning by demonstration [ Kuniyoshi and Inoue, 1993 ] and social learning [ Mataric, 1994 ] is that the other agents is not always friendly nor suggestive to help the agent learn. Rather, the other agents are involved in the task which the agent has to accomplish. That is, they are competitive agents. In this paper, we discuss several issues on the problem of \\learning from other competitive agents" we are facing with in our project. 1 INTRODUCTIO...