基于分层式强化学习的移动机器人导航控制

针对未知环境下的移动机器人导航问题,本文提出了一种基于分层式强化学习的混合式控制方法。利用栅格-拓扑相结合的环境表示及地图学习方法,通过分层式强化学习在不同控制层次的扩展设计移动机器人的反应式和慎思式导航控制,实现了全局导航和局部导航控制的协调优化。实验及测试结果证明,该控制方法能实现导航任务的全局优化,避免陷入局部极小,并对未知动态环境具有较强的适应性。

Mobile Robot Navigation Control Based on Hierarchical Reinforcement Learning

According to the problem of mobile robot navigation in the unknown environment,a hybrid control method based on hierarchical reinforcement learning(HRL) is proposed.Considering the harmonization and optimization of global and local navigation control,the grid-topological map is learned for the environment representation.The grid-topological map is learned for the environment representation to achieve the harmonization and optimization of global and local navigation control.Then reactive and deliberative navigation control of the mobile robot is implemented by extending HRL at different control levels.(1) Reactive control using flat reinforcement learning;(2) Global navigation control by extending reinforcement learning to qualitative state-action space based on topological analysis.Experimental results show that the method can optimize global navigation and avoid getting into local minimum.And it is adaptive to unknown dynamic environments.