基于强化学习算法的多机器人系统的冲突消解策略

Interference Solving Strategy in Multiple Robot System Based on Reinforcement Learning Algorithm

多机器人系统中,随着机器人数目的增加,系统中的冲突呈指数级增加,甚至出现死锁.本文提出了基于过程奖赏和优先扫除的强化学习算法作为多机器人系统的冲突消解策略.针对典型的多机器人可识别群体觅食任务,以计算机仿真为手段,以收集的目标物数量为系统性能指标,以算法收敛时学习次数为学习速度指标,进行仿真研究,并与基于全局奖赏和Q学习算法等其他9种算法进行比较.结果表明所提出的基于过程奖赏和优先扫除的强化学习算法能显著减少冲突,避免死锁,提高系统整体性能.

In a multiple mobile robot system, interface increases exponentially with the increasing number of robots, even deadlock may occur. A reinforcement learning algorithm based on process reward and prioritized sweeping is presented as interference solving strategy. Simulation experiments for forage as task verify the system performance of collected attractors and the learning rate. Comparisons of other nine strategies such as the algorithm based on global reward and Q-learning, show that the presented algorithm based on process reward and prioritized sweeping can decrease interference, avoid deadlock and improve group performance.