摘要
针对强化学习运用于局部路径规划时目标方向不明确易陷入局部最优的情况以及快速探索随机树(RRT)算法规划路径复杂、冗余点多等问题,提出一种融合RRT算法与强化学习(RL)思想的全局路径规划算法。首先,通过RRT全局路径规划算法弱化、减少强化学习算法易于陷入局部最优的问题,并且在一定程度上可以减少规划迭代时间;其次,采用强化学习算法的最大回报奖励机制强化RRT算法在路径规划过程中选择子节点时的目的性,避免过多随机点。实验结果表明,所提算法有效弱化了局部最优所带来的绕远影响,路径长度缩短33.3,凹、凸地形有效节点占比分别提高36.0%和39.6%,侧面反映冗余点数量减少,验证了该算法的可行性。
Aiming at the situation that reinforcement learning to local path planning is not clear about the target direction and easy to fall into local optimality,and the rapidly-exploring random tree(RRT)algorithm has complex planning paths and lots of redundant points,a global path planning integrating RRT algorithm and reinforcement learning(RL)ideas has been proposed.Firstly,the RRT global path planning algorithm is used to weaken and reduce the RL algorithm to avoid falling into the problem of local optimum,which can reduce the planning iteration time to some extent.Secondly,the maximum reward mechanism of the reinforcement learning algorithm is used to strengthen the purpose of the RRT algorithm when selecting child nodes in the path planning process,so as to avoid too many random points.The experimental results suggest that the proposed algorithm weakens the influence of local optimization,shortens the path length by 33.3,increases the proportion of effective nodes in uneven and convex terrain by 36.0%and 39.6%,respectively,reflecting the reduction of redundant points on the side,which verifies the feasibility of the algorithm.
作者
罗国攀
张国良
杨敏豪
LUO Guopan;ZHANG Guoliang;YANG Minhao(School of Automation and Information Engineering,Sichuan University of Science&Engineering,Yibin 644000,China;Artificial Intelligence Key Laboratory of Sichuan Province,Yibin 644000,China)
出处
《四川轻化工大学学报（自然科学版）》
CAS
2024年第2期57-63,共7页
Journal of Sichuan University of Science & Engineering(Natural Science Edition)
基金
四川省应用基础研究项目(2019YJ00413)。
关键词
强化学习
快速探索随机树
回报奖励机制
全局路径规划
reinforcement learning
rapidly-exploring random tree
reward mechanism
global path planning
作者简介
通信作者:张国良(1970-),男,教授,博士,研究方向为先进控制理论、组合导航以及机器人技术,(E-mail)zhgl@sohu.com。
