摘要
针对人工势场法(APF)在避开靠近目标的障碍物时效率低下,易陷入局部极值等问题,提出一种利用强化学习(RL)优化人工势场法的RL-APF避障算法。首先,设计了距离强化因子(DRF)和力强化因子(FRF),通过DRF和FRF将强化学习的奖励功能分解成两部分,当机器人离障碍物太近时,DRF会被激活,而当碰撞发生时,FRF被激活;其次,RL-APF算法通过迭代搜索最佳强化因子的值,以摆脱障碍物并到达目标。实验结果表明,RL-APF算法在简单及复杂环境下各项指标得到不同程度提升,RL-APF算法避障性能优于对比的其他算法,验证了该算法的实用性。
To address the problems that the artificial potential field method(APF)is inefficient in avoiding obstacles close to the target and easy to fall into local extremes,a RL-APF obstacle avoidance algorithm using reinforcement learning(RL)to optimize the artificial potential field method is proposed.Firstly,distance reinforcement factor(DRF)and force reinforcement factor(FRF)are designed to decompose the reward function of reinforcement learning into two parts by DRF and FRF,and DRF is activated when the robot is too close to the obstacle,while FRF is activated when a collision occurs.Then,the RL-APF algorithm iteratively searches for the value of the best reinforcement factor to get rid of the obstacle and reach the target.Finally,the results show that the indexes of RL-APF algorithm are improved in different degrees in simple and complex environments,RL-APF algorithm outperforms the other algorithms compared for obstacle avoidance.Meanwhile,the physical experiments verify the practicality of the proposed algorithm.
作者
张博
周红生
ZHANG Bo;ZHOU Hongsheng(Guoneng Jinjie Energy Co.,Ltd.,Yulin 719319,China;Institute of Acoustics Chinese Academy of Sciences,Beijing 100190,China)
出处
《组合机床与自动化加工技术》
北大核心
2023年第6期7-10,共4页
Modular Machine Tool & Automatic Manufacturing Technique
基金
国家自然科学基金项目(62003249)。
关键词
强化学习
人工势场法
巡检机器人
避障研究
reinforcement learning
artificial potential field method
inspection robot
obstacle avoidance research
