A Distributed Control Method Based on Neighbor Reward for Robot Swarm

Real-time control strategy development remains challenging in the field of controlling multi-robot system. Due to frequent and massive interactions among robots, it is hard to implement a real-time control strategy in the multi-robot system. This phenomenon further limits the development of multi-robot application which adopts the real-time control strategy. To deal with above challenges, this paper proposes a decentralized control policy for multi-robot to avoid collision in an unknown environment. This policy adopts reward value to measure robots' motion and states. With reward value, robots can determine which neighbor robot to communicate with and to learn from, which decreases the interaction cost incurred in the multi-robot system. By the reward, the control strategy of each robot can be optimized through learning from its neighbors. Comparative experiments are conducted in a simulator to evaluate the effectiveness of our policy.