Distributed Soccer Robot Learning And Collaboration

Robocup Robotic Soccer Simulation System is a typical distributed multi-agent system and multi-robot cooperation system. Furthermore, it provides a new standard test-bed and theoretical research model for distributed artificial intelligence. Agent learning and collaboration is hot and difficult in intelligence robot research. Robocup Robotic Soccer System comes down to several robots collaboration, and it is strict to the realtime. It is a challenge to invest the subject of agent learning and collaboration in this environment. This thesis is based on Robocup Soccer System. It mainly focuses on the problem of agent learning and collaboration and explores effective methods to solve this problem.On the aspect of agent study , reinforcement study algorithm need discretize the state spaceand behavior space, but the robot soccer system is a continual space, if discretizing variables results in the dimension disaster, and it can't be realized in the actual system. In this foundation, the improvement algorithm is proposed, only discretization input condition, at the same time, incorporating strengthen study with on-line search enhance the search speed and the efficiency of kicking ball.On the aspect of agent collaboration, in the invironment of Robocup, the system has the character of indeterminacy, so it will effect the agent behavior and the result of collaboration. It is a important problem that how to achieve the collaboration. So, in multi-agent collaboration, It mainly discusses three collaboration model and framework that have been used in the Robocup Robotic Soccer and explores their advantage , disadvantage and feasibility.In the experiments and the research, the algorithms of the reinforcement learning method with neuristic search algorithms gives expression to the advantage ,and the result is satisfied.