基于Hopf振荡器的Spiking-CPG六足机器人步态运动控制

Gait locomotion of hexapod robot based on Spiking-CPGneural system using Hopf oscillator

中枢模式发生器(central pattern generator,CPG)在六足机器人的运动步态控制中起着至关重要的作用。为能够更高效、更低能耗地控制六足机器人的步态运动,提出了一种Spiking-CPG(SCPG)神经网络作为六足机器人的仿生控制系统,其结合Hopf振荡器与以时疏的spiking信号传递信息的仿生神经元LIF(leaky integrate-and-fire),采用环型拓扑结构,使用六组各2000个LIF神经元组成的集合相互连接而成。该SCPG控制系统能够生成六足机器人常见的波动步态、四足步态、三足步态,通过调节相位差参数实现快速、顺滑、稳定的切换运动步态,实时调整所需的频率、振幅,在面对外界干扰时能够在很短的时间内恢复原状,具有很好的鲁棒性。在Webots平台上搭建了一个三维的六足机器人模型,将SCPG的信号输出并经过关节映射函数变换后,来控制六足机器人的运动,验证了所设计六足机器人的运动稳定性和SCPG控制方案的可行性与有效性。最后,在Intel的Loihi芯片上移植了SCPG神经网络控制器,结果表明,其具备高效的执行速度和更低的能耗,在六足机器人的运动控制中具备良好的应用前景。

CPG makes a great contribution to the control of gait motions in hexapod robots.To manage the gait movements of hexapod robots more efficiently and with lower energy consumption,this work introduced a Spiking-CPG(SCPG)neural network as the bionic control system for these robots.The system integrated Hopf oscillators with biologically inspired LIF neurons,which communicated through sparse spiking signals,in a ring topology.It was composed of six groups,each containing 2000 LIF neurons,interconnected to form the network.This SCPG control system could produce common hexapod robot gaits such as the wave,tetrapod,and tripod gaits.It facilitated rapid,smooth,and stable transitions between motion gaits by adjusting phase difference parameters,and could dynamically adjust the required frequency and amplitude,exhibiting excellent robustness by quickly recovering from external disturbances.It controlled a simple 3D hexapod robot model on the Webots platform.It deve-loped the robot by transforming the SCPG signal output through joint mapping functions,validating the motion stability of the designed hexapod robot and the feasibility and effectiveness of the SCPG control scheme.Finally,it ported the SCPG neural network controller proposed to Intel’s Loihi chip,showing that it has high execution speed and lower energy consumption,indicating a promising prospect for application in the motion control of hexapod robots.