基于RPkNN-Sarsa(λ)强化学习的机器人路径规划方法

摘要

The method of robot path planning based on kNN-Sarsa(λ) reinforcement learning has fast convergence speed, but the algorithm is easy to fall into local optimal value and does not consider incomplete observability of environmental information. With regards to this, this paper designed a method of random perturbation kNN-Sarsa(λ) reinforcement learning algorithm. Also, it processed sensors detection data uncertainty using Bayesian theory. In addition, it used grid map to establish simulation environment model. The simulation experimental results show that the method not only has rapid convergence speed by alleviating the local optimal problem of kNN-Sarsa(λ) algorithm, but also can find a shortest path in the case of sensors detection data uncertainty.%基于kNN-Sarsa(λ)强化学习的机器人路径规划方法虽然收敛速度快,但该算法容易陷入局部最优值,且未考虑环境信息的不完全可观测性.为此,设计了一种随机扰动(random perturbation) kNN-Sarsa(λ)强化学习算法,利用Bayesian规则对传感器探测信息的不确定性进行了处理,建立了基于栅格地图的仿真环境模型.仿真实验结果表明,该方法不仅收敛性好,能有效缓解kNN-Sarsa(λ)算法易陷入局部最优的现象,且在传感器探测信息不确定的情况下仍能搜索到最优路径.