China Mechanical Engineering

Previous Articles     Next Articles

An Identification Method of Hybrid Dynamics System for Biped Robots Based on Neural Network

WU Xiaoguang1;ZHANG Tianci1;WEI Lei1;LI Yanhui1;WANG Tingjin1;ZHANG Bo2   

  1. 1.Key Lab of Measurement Technology and Instrumentation of Hebei Province,Yanshan University,Qinhuangdao,Hebei,066004
    2.CRRC Tangshan Co.,Ltd.,Tangshan,Hebei,063000
  • Online:2018-07-25 Published:2018-07-27
  • Supported by:
    National Natural Science Foundation of China (No. 61503325)
    Hebei Provincial Natural Science Foundation of China (No. F2014203246)
    China Postdoctoral Science Foundation(No. 2015M581316)

一种基于神经网络的双足机器人混合动力学系统辨识方法

吴晓光1;张天赐1;韦磊1;李艳会1;王挺进1;张2   

  1. 1.燕山大学河北省测试计量技术及仪器重点实验室,秦皇岛,066004
    2.中国唐山机车车辆有限公司,唐山,063000
  • 基金资助:
    国家自然科学基金资助项目(61503325);
    河北省自然科学基金资助项目(F2014203246);
    中国博士后科学基金资助项目(2015M581316)
    National Natural Science Foundation of China (No. 61503325)
    Hebei Provincial Natural Science Foundation of China (No. F2014203246)
    China Postdoctoral Science Foundation(No. 2015M581316)

Abstract: In view of difficulties in identification of hybrid dynamics system for biped robots,identification conditions for continuous and discrete hybrid system in terms of asymptotic stability were analyzed and deduced.And then,a joint identification method was proposed based on combination of radial basis function neural network optimized by chaos particle swarm optimization and dynamic fuzzy neural network.Radial basis function neural network which was optimized by chaos particle swarm optimization to identify continuous swing stages,and dynamic fuzzy neural network was used for discrete collision stages.According to the same variable's coupling and transition relationship during two phases,the accurate identification for hybrid systems of biped robots was realized.The simulation results show that,the identified and predicted results are of high precision in this method.

Key words: biped robot, hybrid dynamics system, system identification, neural network

摘要: 针对双足机器人的混合动力学系统辨识问题,从系统渐进稳定性角度分析,推导出连续与离散混合系统的可辨识条件,提出了一种基于混沌粒子群优化的径向基函数神经网络与动态模糊神经网络的联合辨识方法。利用混沌粒子群优化的径向基函数神经网络辨识双腿的连续摆动阶段,利用动态模糊神经网络辨识离散的足地碰撞阶段;依据两阶段同一变量的耦合、转换关系,实现了对双足机器人整体混合系统的准确辨识。仿真实验结果表明,该方法辨识和预测结果具有较高的准确度。

关键词: 双足机器人, 混合动力学系统, 系统辨识, 神经网络

CLC Number: