具有未知侧滑和打滑的WMR强化学习自适应神经网络控制  

Reinforcement learning adaptive neural network control of WMR with unknown skidding and slipping

作　　者：叶锦华[1] 吴海彬[1] 

机构地区：[1]福州大学机械工程及自动化学院,福建福州350116

出　　处：《福州大学学报（自然科学版）》2016年第2期219-224,共6页Journal of Fuzhou University(Natural Science Edition)

基　　金：国家自然科学基金资助项目(51175084);福建省自然科学基金资助项目(2015J05121);福州大学科研启动基金资助项目(510078);福州大学科技发展基金资助项目(650053)

摘　　要：利用反演设计,提出一种强化学习自适应神经网络轮式移动机器人(WMR)轨迹跟踪控制方法.首先在极坐标下建立WMR的轨迹跟踪误差模型,并基于此设计运动学控制器.然后,针对WMR动力学系统,设计自适应神经网络控制器.结合强化学习机制,同时对系统未知侧滑、打滑和模型不确定性进行优化补偿,并引入鲁棒控制项来消除补偿误差的影响,进一步提高了控制效果.所提控制方法使得闭环系统稳定,且最终一致有界收敛,其有效性通过数值仿真结果得到了验证.A reinforcement learning adaptive neural network trajectory tracking control scheme is proposed for WMR,based on back stepping technique. Firstly,the trajectory tracking error model is established,and the kinematic controller is designed based on this model. Then,for WMR dynamic system,the adaptive neural network controller with reinforcement learning is designed,and unknown skidding,slipping and model uncertainties of the system are compensated optimally,the robust compensators are also used to eliminate the effects of compensating error,so the control performance is enhanced. The stability and ultimately uniformly bounded convergence of system are guaranteed with proposed control scheme. Simulations prove the validity of the proposed control scheme.

关 键 词：轨迹跟踪 自适应神经网络 强化学习控制 非完整轮式移动机器人 不确定系统 

分 类 号：TP24[自动化与计算机技术—检测技术与自动化装置]