基于路径规划和深度强化学习的机器人避障导航研究
ROBOT OBSTACLE AVOIDANCE NAVIGATION BASED ON PATH PLANNING AND DEEP REINFORCEMENT LEARNING
康振兴1
作者信息
- 1. 合肥工业大学仪器科学与光电工程学院 安徽合肥 230009
- 折叠
摘要
针对移动机器人的长距离避障导航问题,提出结合深度强化学习(Deep Reinforcement Learning,DRL)和路径规划(Path Planning,PL)的避障导航算法.该方法通过快速扩展随机树(Rapidly Exploring Random Tree,RRT)算法在长距离的路径上进行规划,根据生成的路径节点,将长距离路径划分为若干短距离,而在短距离的导航问题上利用深度强化学习的算法,训练一个具有环境感知和智能决策能力的端到端避障导航模型.仿真实验表明,相较于仅用DRL的避障导航,该方法使移动机器人的长距离避障导航性能有了大幅度提升,解决了 DRL在长距离避障导航任务上的局限性问题.
Abstract
Aimed at the long-distance obstacle avoidance navigation problem of mobile robots,an obstacle avoidance navigation algorithm combining deep reinforcement learning(DRL)and path planning(PL)is proposed.This method used the rapidly exploring random tree(RRT)algorithm to plan the long-distance path.According to the generated path nodes,the long-distance path was divided into several short distances.And for the navigation problem in the short distance,DRL algorithms were used to train an end-to-end obstacle avoidance navigation model with environmental perception and intelligent decision-making capabilities.Simulation experiments show that compared with obstacle avoidance navigation using only DRL,this method greatly improves the long-distance obstacle avoidance navigation performance of mobile robots,and solves the limitations of DRL in long-distance obstacle avoidance navigation tasks.
关键词
深度强化学习/路径规划/移动机器人/长距离避障导航Key words
Deep reinforcement learning/Path planning/Mobile robot/Long-distance obstacle avoidance navigation引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据