基于约束型TD3的动态探索噪声改进算法
(1.太原师范学院 计算机科学与技术学院,山西 晋中 030619;2.太原工业学院,山西 太原 030008)
摘要
本研究针对无约束探索可能致使移动小车受损的问题,提出一种融合自适应噪声探索与拉格朗日乘子约束的强化学习方法,旨在优化小车到达目标点的轨迹规划。此方法通过动态调节噪声来提升探索效率,运用TD3算法应对连续动作空间,借助拉格朗日乘子法处理约束条件,这与直接在马尔科夫决策过程(MDP)中添加非期望行为惩罚的方式有所不同。仿真实验表明,该方法能够有效地引导小车避开障碍物,减少违反约束的情形,同时保障任务的安全性与可靠性,呈现出良好的训练收敛特性。
关键词
安全强化学习;约束马尔科夫决策过程;轨迹规划;TD3算法
参考文献
- [1] 计时鸣,黄希欢.工业机器人技术的发展与应用综述 [J].机电工程,2015,32(1):1-13.
- [2] 王雪松,王荣荣,程玉虎.安全强化学习综述 [J].自动化学报,2023,49(9):1813-1835.
- [3] MORALES E F,ZARAGOZA J H. An Introduction to Reinforcement Learning [M]//SUCAR L E,MORALES E F,HOEY J. Decision Theory Models for Applications in Artificial Intelligence:Concepts and Solutions [S.I.]:IGI Global,2012:63-80.
- [4] ACHIAM J,HELD D,TAMAR A,et al. Constrained Policy Optimization [C]//Proceedings of the 34th International Conference on Machine Learning.Sydney:PMLR,2017,70:22-31.
- [5] SUI Y,GOTOVOS A,BURDICK J,et al. Safe Exploration for Optimization with Gaussian Processes [C]//International Conference on Machine Learning.Lille:PMLR,2015:997-1005.
- [6] LILLICRAP T P,HUNT J J,PRITZEL A,et al. Continuous Control with Deep Reinforcement Learning [J/OL].[2024-10-03].arXiv:1509.02971 [cs.LG].https://doi.org/10.48550/arXiv.1509.02971.
- [7] STOOKE A,ACHIAM J,ABBEEL P. Responsive Safety in Reinforcement Learning by PID Lagrangian Methods [J/OL].arXiv:2007.03964 [math.OC].[2024-10-06].https://doi.org/10.48550/arXiv.2007.03964.
- [8] IBARZ J,TAN J,FINN C,et al. How to Train your Robot with Deep Reinforcement Learning: Lessons we have Learned [J].The International Journal of Robotics Research,2021,40(4-5):698-721.
- [9] MARCHESINI E,CORSI D,FARINELLI A. Benchmarking Safe Deep Reinforcement Learning in Aquatic Navigation [C]//2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).Prague:IEEE,2021:5590-5595.
- [10] 李潇宇,张君华,郭晓光,等.基于强化学习的机器人底盘能量管理与路径规划优化算法 [J].农业工程学报,2024,40(21):175-183.
- [11] 聂振邦,于海斌.基于参考轨迹的移动机器人避碰决策及轨迹跟踪方法 [J].计算机集成制造系统,2023,29(9):2879-2889.
- [12] ISCEN A,YU G,ESCONTRELA A,et al. Learning Agile Locomotion Skills with a Mentor [C]//2021 IEEE International Conference on Robotics and Automation (ICRA).Xi'an:IEEE,2021:2019-2025.
- [13] DALAL G,DVIJOTHAM K,VECERIK M,et al. Safe Exploration in Continuous Action Spaces [J/OL].arXiv:1801.08757 [cs.AI].[2024-10-10].https://doi.org/10.48550/arXiv.1801.08757.
- [14] RAMEZANI M,HABIBI H,SANCHEZ-LOPEZ J L,et al. UAV Path Planning Employing MPC-Reinforcement Learning Method Considering Collision Avoidance [C]//2023 International Conference on Unmanned Aircraft Systems (ICUAS).Warsaw:IEEE,2023:507-514.
- [15] JI J,ZHOU J,ZHANG B,et al. Omnisafe: An Infrastructure for Accelerating Safe Reinforcement Learning Research [J].Journal of Machine Learning Research,2024,25(285):1-6.
- [16] ZHAO W Y,HE T N,CHEN R,et al. State-wise Safe Reinforcement Learning: A Survey [J/OL].arXiv:2302.03122 [cs.LG].[2024-0-16].https://doi.org/10.48550/arXiv.2302.03122.
- [17] FUJIMOTO S,HOOF H,MEGER D. Addressing Function Approximation Error in Actor-Critic Methods [J/OL].arXiv:1802.09477 [cs.AI].[2024-10-11].https://doi.org/10.48550/arXiv.1802.09477.
作者简介
陈春甫(1999—),男,汉族,山西晋中人,硕士研究生在读,研究方向:强化学习、机器人系统;通信作者:穆煜(1978—),男,汉族,山西太原人,讲师,博士,研究方向:机器人运动控制与规划;韩凯涛(1999—),男,汉族,山西吕梁人,硕士研究生在读,研究方向:光电信息处理与通信技术。