Technical Issues

In addition to issue in the task-I and -II, three issues are left.

• Since the control architecture is not centralized but decentralized, each agent should know capabilities in passing and receiving skills of not only itself but the other. That is, the agent should estimate the level of the other agent skills. This means agent modeling. The problem includes partial observability due to perceptual limitations.
• Even though both agents have reasonable passing and receiving skills, the timing of passing and receiving must be coordinated. If both agents try to learn to improve their own behavior independently, the learning may not converge because the policy of the other changes simultaneously. To prevent this situation, one of the agents should be a coach (fixed policy) and the other, a learner. In this case, modeling of the learner is another issue for good teaching.
• In all these behaviors, and especially in a pass interception, the success rate will drastically increase if the agent can predict the ball holder's ball controlling behavior, specifically, when and in which direction it will kick. For this, the agent should first find a ball holder, track its motion, and predict the oncoming event based on the relative positions of the ball and the surrounding agents, such as the potential pass receivers. A primitive example of vision based interception of a static obstacle from another robot's trajectory has been demonstrated [Kuniyoshi, 95]. However, a general interception in fully dynamic situations like soccer playing is an open problem.
• Selection of passing direction depends on the motions of opponents. This introduces the opponent modeling issue which makes the cooperative behavior much harder to realize.