A reinforcement learning method to improve the sweeping efficiency for an agent

This article proposes a reinforcement learning method aimed at improving the sweeping efficiency of an agent. In the proposed method, an agent attempts to avoid overlapping a swept field by using a combination of distances from the agent to obstacles and the information which expresses whether a field in front of the agent has already been swept. We carried out several computer simulations to evaluate basic characteristics and performance of the proposed method. The empirical results showed that the agent behaves effectively in the field compared to an agent with fixed heuristics.