Biorobotics Laboratory BioRob

Stéphane Mojon (Diploma project 2004)

Using nonlinear oscillators to control the locomotion of a simulated biped robot

After millions years of evolution, human beings are able to walk, to run and to jump with an incredible agility. Therefor, one could think that making biped robot walk would be simple to do. But it turned out that the problem is more complex than it seams at first sight. Many complex models, based on the exact robot's dynamics, have been developed. But none of them are able to deal with natural and unknown environments.

So in this project another and bio-inspired approach has been tried. The controller used was directly inspired by the Central Pattern Generator (CPG) observed in some simple vertabrate. Then the unknown parameters of this controller have been optimized using a genetic algorithm.

With this approach we were able to make a simulated biped robot walk at a speed of 1.5 km/h. Unfortunately this walk was not robust against external perturbations as we initially expected. However the evolved CPG had some interesting properties: It has the ability to stabilized itself from any initial conditions and its output signals can be dynamically and smoothly changed.