一类用于井下路径规划问题的Dyna_Q学习算法

A Dyna_Q-learning Algorithm Used in Underground Path Planning

在基于目标的强化学习任务中,欧氏距离常用于Dyna_Q学习的启发式规划中,但对于井下救援机器人路径规划这类状态空间在欧氏空间内不连续的任务效果不理想。针对该问题,文章引入流形学习中计算复杂度较低的拉普拉斯特征映射法,提出了一种基于流形距离度量的改进Dyna_Q学习算法,并在类似于井下环境的格子世界中进行了仿真研究。仿真结果验证了该算法的有效性。

The Euclidean distance is usually used in heuristic planning of Dyna_Q-learning based on reinforcement learning tasks of goal position. But it is not suitable for these tasks whose state space is not continuous in Euclidean space such as path planning of disaster rescue robot in underground coal mine. For the problem, the paper introduced the Laplacian Eigenmap whose computational complexity is lower in manifold learning, then proposed an improved Dyna_ Q-learning algorithm based on manifold distance metric. The proposed algorithm is simulated in grid world that is similar to underground environment. The simulation results verified validity of the algorithm.