基于FBFN的鲁棒仿生学习系统设计及在运动平衡控制中的应用  

Robust Bionic Learning System Design Based on FBFN and Its Application to Motion Balance Control

作　　者：蔡建羡[1,2] 阮晓钢[1] 

机构地区：[1]北京工业大学电子信息与控制工程学院,北京100124 [2]防灾科技学院,河北廊坊065201

出　　处：《机器人》2010年第6期732-740,共9页Robot

基　　金：国家自然科学基金资助项目(60774077);国家863计划资助项目(2007AA04Z226);北京市教委重点项目(KZ200810005002)

摘　　要：针对两轮直立式机器人的运动平衡控制问题,结合OCPA仿生学习系统,基于模糊基函数,设计了一种鲁棒仿生学习控制方案.它不需要动力学系统的先验知识,也不需要离线的学习阶段.鲁棒仿生学习控制器主要包括仿生学习单元、增益控制单元和鲁棒自适应单元3部分.仿生学习单元由模糊基函数网络(FBFN)实现,FBFN不仅执行操作行为产生功能,逼近动力学系统的非线性部分,同时也执行操作行为评价功能,并利用性能测量机制提供的误差测量信号,产生取向值信息,对操作行为产生网络进行调整.增益控制单元的作用是确保系统的稳定性和性能,鲁棒自适应单元的作用是消除FBFN的逼近误差及外部干扰.此外,由于FBFN的参数是基于李亚普诺夫稳定性理论在线调整的,因此进一步确保了系统的稳定性和学习的快速性.理论上证明了鲁棒仿生学习控制器的稳定性,仿真实验结果验证了其可行性和有效性.Aiming at the motion balance control problem of a two-wheeled upright robot and combining OCPA（operant conditioning probabilistic automaton） bionic learning system,a robust bionic learning control scheme based on fuzzy basis function is designed.It doesn＇t require prior knowledge of dynamic system or off-line learning phase.The architecture of the robust bionic learning controller contains a bionic learning unit,a gain controller unit and a robust adaptive control unit The bionic learning unit is realized based on fuzzy basis function network（FBFN）,and it is not only employed to generate operant action for approximating nonlinear parts,but also to evaluate the operant action.Based on the error measurement signal provided by performance measurement mechanism,the orientation information is generated to tune operant action generation network.The function of the gain controller unit is to guarantee the stability and performance of system,and the function of the robust adaptive unit is to eliminate the approximation error of the FBFN and external disturbances.Besides, the proposed scheme can significantly shorten the learning time and guarantee the stability of system by on-line tuning all parameters of FBFN based on Lyapunov stability theory.The stability of the robust bionic learning controller is proved in theory,and its feasibility and effectiveness can be demonstrated from the results of simulation experiment.

关 键 词：仿生学习 模糊基函数网络 鲁棒 运动平衡控制 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]