Locomotion Control of a Six-Legged Walking Robot Based on Central Pattern Generator Network

Abstract

In recent years, much attention has been given to bionic control. The motion control of legged robots based on central pattern generator (CPG) has become one of the important branches of bionic control field as a research hotspot. CPG can produce steady rhythm without any high-level external control signal and sensory feedback, which does not require the model process on environment. The objective of this thesis is to control the locomotion of a six-legged walking robot based on central pattern generator network. To do this task, the following problems are considered. Firstly, a six-legged walking robot is presented. Secondly, wave gait, quadruped gait and tripod gait should be generated by CPG network. Thirdly, mapping function is designed to map the output signal of CPG network to the workspace trajectories of the corresponding legs. Fourthly, instead of using inverse kinematics to calculate joint angles, a differential kinematics algorithm is applied for the end effector to follow the trajectory generated by CPG network. The following tasks are done to solve these problems. First, a six legged robot platform is developed with several interconnected devices such as AX-12 servomotor, DSP microcontroller, IMU sensor, etc. The kinematics of one leg of the six-legged robot is presented, and the Denavit-Hartenberg (DH) convention is adopted to define the modeling parameters which allow the construction of the forward kinematics function by composition of the individual coordinate transformation. Second, a CPG model based on Kimura’s neural oscillators is proposed. To get a suitable control signal of CPG, the parameters of CPG model are analyzed based on a single-parameter-analysis method using simulation. Third, a CPG network to generate control signal for wave gait, quadruped gait and tripod gait using six CPG models is built. Fourth, mapping functions to change the control signal of CPG into the workspace trajectory are proposed. For swing phase and retract phase of gait, different mapping functions are used. The end effector trajectory of each leg is given as mapping functions of the output of signal of the corresponding oscillator. Fifth, to realize the leg end effector follow the trajectory obtained by the mapping functions, a differential kinematics algorithm is applied to solve inverse kinematics problem easily, and the differential kinematics is represented as a linear mapping between the joint velocity space and the operational velocity space. Finally, simulation and experimental results are done to demonstrate the effectiveness of the proposed controllers.