动态约束下可重构模块机器人分散强化学习最优控制  

Decentralized reinforcement learning optimal control for time varying constrained reconfigurable modular robot

作　　者：董博[1] 刘克平[2] 李元春[2] 

机构地区：[1]吉林大学控制科学与工程系,长春130022 [2]长春工业大学控制工程系,长春130012

出　　处：《吉林大学学报（工学版）》2014年第5期1375-1384,共10页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(61374051;60974010);吉林省科技发展计划项目(20110705)

摘　　要：基于ction-critic-identifier(ACI)与RBF神经网络,提出了一种外界动态约束下的可重构模块机器人分散强化学习最优控制方法,解决了存在强耦合不确定性的模块机器人系统的连续时间非线性最优控制问题。文中将机器人动力学模型描述为一个交联子系统的集合,基于连续时间MDPs性能指标,结合ACI与RBF神经网络,对子系统最优值函数,最优控制策略及总体不确定项进行辨识,使系统满足HJB方程下的最优条件,从而使可重构模块机器人子系统渐进跟踪期望轨迹,跟踪误差收敛且有界。采用Lyapunov理论对系统稳定性进行证明,数值仿真验证了所提出的分散控制策略的有效性。Based on Action-Critic-Identifier （ACI） and Radial Basis Function （RBF） neural network, a novel decentralized reinforcement learning optimal control method for time varying constrained reconfigurable modular robot is presented. The continuous time nonlinear optimal control problem of strongly coupled uncertainty robotic system is solved. The dynamics of the robot is described as a synthesis of interconnected subsystems. As a precondition to the continuous-time MDPs performance indicators, the optimal value function, optimal control policy and global uncertainty of the subsystems are estimated combing with ACI and RBF network. The optimal conditions of HJB equation with regard to the subsystem are satisfied, so that the reconfigurable modular robot system can track the desired trajectory in a short time and the estimation error can converge to zero in finite time. The stability of the system is confirmed by Lyapunov theory. Simulations are performed to illustrate the effectiveness of the proposed decentralized control scheme.

关 键 词：自动控制技术 可重构模块机器人 强化学习 非线性最优控制 分散控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]