Sharing learning policies between multiple mobile robots

Learning of a complex task usually requires a long learning period. In order to reduce the time of learning, the task is divided into several subtasks. Multiple agents can be used to serve a complex task by learning these subtasks concurrently. With a good knowledge sharing mechanism, the learning policy can be shared or exchanged among these agents and can enhance their learning efficiency. The learning policy is a mapping from system states to actions. The mechanism of sharing or exchanging learning knowledge among multiagent system is proposed. An index of expertise, which indicates the skill level of each learning agent, is presented. This index is used to select the best preferable advice among multiple advices, which can increase the probability of finding solution in the search space. The experiment in which the learning knowledge is exchanged between a mobile robot and a computer simulated agent is implemented in order to verify the validity of the proposed algorithm. The experimental results show that the learning efficiency of the advisor agent is increased and the advisee robot can use the given advice for avoiding collision with obstacle successfully in the real world implementation.