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

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.