基于脉冲推力的半被动双足机器人无模型神经网络控制

摘要

This paper studied the stable walking control of quasi-passive biped robots on level ground.Firstly,it used the simplest point-foot model as the dynamical model of the biped robot and applied an impulsive push to the stance leg preferably immediately before heel strike.Secondly,in view of the nonlinear characteristics of the biped robot model,it introduced the functional link artificial neural network (FLANN) control algorithm based on trigonometric function polynomials into the biped robot system,which was used to generate the toe-off impulse.Then it used the modal-free simultaneous perturbation stochastic approximation (SPSA) algorithm based on data driven to update the weights of the FLANN.Thirdly,it used Poincaré map to analyze the stability condition for the biped robot walking.Finally,based on theoretical analysis,it carried out computer simulations to validate the effectiveness of the proposed algorithm.Simulation results show that the proposed algorithm has better convergence performance than the iterative learning algorithm (ILC),and all the eigenvalues of the Jacobian matrix are inside the unit circle,so it satisfies the stability condition.%研究了半被动双足机器人的平面稳定行走控制问题.以最简行走模型为动力学模型,采用沿支撑腿方向的脚后跟脉冲推力作为行走动力源.考虑到系统模型的非线性特征,将基于三角函数扩展的函数链接型人工神经网络控制算法引入到机器人系统中,以产生系统所需的脉冲推力.采用基于数据驱动的无模型同步扰动随机逼近算法对神经网络的权值进行更新,利用庞加莱映射方法分析了半被动双足机器人行走的稳定条件.在理论分析的基础上,对该算法进行了仿真研究.仿真结果表明,算法在收敛快速性上要优于迭代学习控制算法,可以实现双足机器人平面上的稳定周期行走,且雅可比矩阵的特征值均位于单位圆内,满足系统的稳定条件.