Dynamic Walkng of Legged Machines

TL;DR

This paper presents a simple, control-splitting approach based on the Linear Inverted Pendulum model to achieve stable, natural bipedal walking in simulation, reducing complexity and computational costs.

Contribution

It introduces a novel control system that simplifies legged locomotion by splitting control into height, velocity, and balance, using linear foot placement based on body velocity.

Findings

Achieved stable walking over 200 steps in simulation

Walk cycle is cyclic, stable, and natural

Control system reduces complexity and computational costs

Abstract

Locomotion of legged machines faces the problems of model complexity and computational costs. Algorithms based on complex models and/or reinforcement learning exist to solve the walking control task. In this project, we aim to develop a bipedal walking control system based on a simple model the Linear Inverted Pendulum model. In order to simplify the complex process of controlling legged locomotion, we make use of the technique of splitting the control into three parts as height control, forward velocity control and balance control. The forward velocity of the body has a linear relationship with the foot placement, therefore we use a linear function to realise foot placement. Our control system achieves stable walking gait in a simulated environment, where our bipedal robot walks more than 200 steps with a cyclic pattern in a stable, dynamic and almost natural manner. The experimental data are presented and analysed.