Real-time autonomous distributed cooperation under dynamically changing environments

我们研究了实时多智能体系统，在该系统中，一些自治的智能体只能共享关于其动态变化的环境的不完全信息，但必须通过在真实的时间内相互合作来共同实现其目标。我们选择RoboCup soccer和RoboCupRescue作为研究的实验平台，是因为这两个平台最适合研究实时分布式协作系统的逻辑和软件方面，我们的研究成果最显著的一点是，所开发的智能体能够尽可能地以类人的方式解决这些复杂的实时分布式协作问题。例如，我们的足球智能体只使用类似人类的短喊来进行通信，以使战术团队发挥作用。尽管有这样的限制，我们的代理团队赢得了这一时期在日本举行的所有正式比赛。我们还开发了有效的可视化调试环境，代理开发。从编程的角度来看，这种代理开发只不过是并发编程工作，这是最难测试和调试的，因为不会有相同的情况重复存在其他冒犯对手。

We investigated on real-time multi-agent systems in which a number of autonomous agents that can share only incomplete information about their dynamically changing environment but have to achieve their objective collectively by cooperating with each other in real time. We chose as the research test-beds RoboCup soccer and RoboCupRescue because these are best suited for research on logical and software aspects of real-time distributed cooperative systems.The most remarkable point of our research results is that developed agents manage to solve these complicated real-time distributedcooperation problems in humanlike ways as far as possible. For example, our soccer agents use only humanlike short shouts for their communication to make tactical team play. In spite of this limitation, our agent teams won all official competitions in Japan held in this term.We also developed effective visual debugging environments for agent development. Such agent development is, from the programming viewpoint, nothing other than concurrent programming effort, which is most difficult to test and debug because there will be no same situations repeated with the existence of other offending opponents.