An Exploration Of Biologically-inspired Humanoid Walking

Humanoid robot is a complex multi-rigid-body biped walking system, which hasmultiple-input multiple-output, strong coupling, nonlinear and other characteristics.This not only requires high performance of planning algorithm using to achieve stablewalking, but also requires its motion control system having high performance.The biological inspiration can endow a robot with some level of intelligence insome context by the simulation of biological processes and nature phenomena. Aftersummarized the gait planning and bio-inspired motion control strategies of thehumanoid robot, this paper analyzes the general procedures and biological principle ofthe human walking, then has proposed a interactive learning control model. Finally, weuse the bio-inspired control strategy based on Cerebella Model Articulation Controllerand interactive learning controller based on GCMAC neural network to achieve thewalking of the humanoid robot.In order to achieve human walking we built the coordinates firstly. The mainprocess of walking control of the robot can be taken as the switch process of multi-basecoordinate system, and the switch matrix was found. The movement of center mass inthe ground base coordinate system was planned then, and the robot single leg movementis divided into three states, which are individual support, joint support and non support.The movement of individual support state can be got from the movement in the groundbase coordinate system; The movement of joint support state and non support state canbe got from the Bezier curve whose boundary conditions was set. By using ZMPprinciples, the stability of gait was proven. Followed, we design the CMAC-PD parallelcontroller and the interactive learning controller based on the GCMAC neural networkto achieve the gait control of the humanoid robot.Finally, we simulate the humanoid robot walking by using the humanoid robotsimulation platform which is built by ODE physics engine. It is not only proved that it isreasonable to plan the robot gait by using the gait planning algorithm based onmulti-base coordinates, but also proved that the CMAC neural network based onbio-inspired control strategy which is used to the gait control can reduce the robotwalking error and achieve the gait online learning and adjusting. Meanwhile, therationality of the interactive learning control model has also been proved.