摘要
提出了一种新的不确定机器人跟踪控制策略.在计算力矩结构的基础上引入一个层叠小脑模型(CMAC)补偿控制项,利用层叠结构CMAC分层学习的特性调整网络泛化和逼近能力,并从理论上分析了网络的收敛性.为了确保系统误差一致最终有界收敛,分别设计了粗/细子网的权值更新律.最后,在网络学习稳定的基础上,采用自适应鲁棒项抵消网络最终学习误差.与传统计算力矩法相比,在不要求加速度测量和惯性矩阵求逆的情况下,算法给出清晰的跟踪误差收敛域.基于6自由度并联机器人的仿真实例验证了算法的有效性.
A new kind of uncertain robot tracking control strategy is presented. A cascaded cerebellar model articulation controller (CMAC) compensation term is introduced based on computed torque structure. The delaminated learning property of cascaded CMAC is used to amend the net's generalization and approach ability, and the convergence of cascaded CMAC is analyzed theoretically. To insure the system error is uniformly ultimately bounded (UUB), the update laws of the coarse/fine subnet are designed respectively. At last the adaptive robust term is added to compensate the unknown ultimately learning error when the net become stable. Compared with the computed torque method, the algorithm can give error convergence bound clearly without acceleration measurement and inverse inertia matrix. The performance of the algorithm is verified through the simulation studies on 6 degrees of freedom parallel manipulator.
出处
《西安交通大学学报》
EI
CAS
CSCD
北大核心
2003年第6期569-572,共4页
Journal of Xi'an Jiaotong University
关键词
计算力矩
层叠小脑模型
一致最终有界
并联机器人
Computer simulation
Convergence of numerical methods
Learning systems
Lyapunov methods
Neural networks
Systematic errors