Learning to swim: a dynamical systems approach to mimicking fish swimming with CPG
Learning to swim: a dynamical systems approach to mimicking fish swimming with CPG
复制标题
学习游泳：用 CPG 模仿鱼游泳的动力系统方法
DOI：
10.1017/s0263574712000343
复制
发表时间：
2012-07
期刊：
影响因子：
2.7
通讯作者：
J. Liang
中科院分区：
文献类型：
--
作者：
T. Wang;Y. Hu;J. Liang
摘要中央模式发生器（Central Pattern Generators，CPG）可以产生鲁棒的、平滑的和协调的振荡信号，用于具有多个自由度的机器人的运动控制，但是针对期望的运动模式的CPG参数的调谐构成了非常困难的任务。本文针对这一问题，提出了一种基于自适应CPG网络的多关节机器鱼仿鱼游动步态生成方法。我们的方法将相关的CPG参数转换为动态系统，作为CPG网络动态的一部分。为了再现游动鱼的身体运动，我们使用轨迹近似方法计算的关节角作为CPG网络的教学信号，该网络被建模为一个耦合的Hopf振子链。提出了一种新的耦合方案，以消除传入信号对振荡器幅度的影响。利用振子的相空间表示，建立了固有频率、耦合权重和振幅的学习规则。通过自适应机制，CPG网络可以对示教信号的频率、幅度和相位关系进行编码。由于霍普夫振子表现出极限环行为，学习的运动模式对扰动是稳定的。此外，由于CPG模型的非线性特性，可以以平滑的方式进行目标行进体波的修改。数值实验验证了所提出的学习规则的有效性。
SUMMARY Central Pattern Generators (CPGs) can generate robust, smooth and coordinated oscillatory signals for locomotion control of robots with multiple degrees of freedom, but the tuning of CPG parameters for a desired locomotor pattern constitutes a tremendously difficult task. This paper addresses this problem for the generation of fish-like swimming gaits with an adaptive CPG network on a multi-joint robotic fish. Our approach converts the related CPG parameters into dynamical systems that evolve as part of the CPG network dynamics. To reproduce the bodily motion of swimming fish, we use the joint angles calculated with the trajectory approximation method as teaching signals for the CPG network, which are modeled as a chain of coupled Hopf oscillators. A novel coupling scheme is proposed to eliminate the influence of afferent signals on the amplitude of the oscillator. The learning rules of intrinsic frequency, coupling weight and amplitude are formulated with phase space representation of the oscillators. The frequency, amplitudes and phase relations of the teaching signals can be encoded by the CPG network with adaptation mechanisms. Since the Hopf oscillator exhibits limit cycle behavior, the learned locomotor pattern is stable against perturbations. Moreover, due to nonlinear characteristics of the CPG model, modification of the target travelling body wave can be carried out in a smooth way. Numerical experiments are conducted to validate the effectiveness of the proposed learning rules.
登录
查看更多内容
影响因子：
20.6
作者：
Y. Terada;I. Yamamoto
通讯作者：
Y. Terada;I. Yamamoto
影响因子：
3.7
作者：
LIGHTHILL, MJ
通讯作者：
LIGHTHILL, MJ
影响因子：
4
作者：
Righetti, Ludovic;Buchli, Jonas;Ijspeert, Auke Jan
通讯作者：
Ijspeert, Auke Jan
影响因子：
6
作者：
Daibing Zhang;D. Hu;Lincheng Shen;Haibin Xie
通讯作者：
Daibing Zhang;D. Hu;Lincheng Shen;Haibin Xie
DOI：
10.1109/icit.2005.1600829
发表时间：
2005-12
期刊：
2005 IEEE International Conference on Industrial Technology
影响因子：
--
作者：
Long Wang;Shuo Wang;Zhiqiang Cao;M. Tan;Chao Zhou;Haiquan Sang;Zhizhong Shen
通讯作者：
Long Wang;Shuo Wang;Zhiqiang Cao;M. Tan;Chao Zhou;Haiquan Sang;Zhizhong Shen