Biorobotics Laboratory BioRob

Print

Julien Nicolas (Diploma Project 2005)
Supervisors: Ludovic Righetti and Auke Ijspeert



Artificial evolution of controllers based on non-linear oscillators for bipedal locomotion


Millions of years of evolution made humans able to walk and run with an incredible precision. Although bipedal locomotion could seem to be a simple mechanical problem, researches done on artificial locomotion proved that it is far from obvious to generate a walking gait that is as fast and stable as humans do.
This work is directly inspired from biological considerations by using CPG based controllers and artificial evolution. First, several models of CPGs, implemented on a simulated robot, were explored using genetic algorithms to find which system is best suited to locomotion. After that, our model was improved by adding feedback pathways in the CPGs in order to be able to modify the speed of locomotion.

The approach described above allowed us to make the robot walk at approximatively 0.28 m/s without feedback. We also managed to increase the system's frequency by adding feedback pathways which allowed us to increase the velocity by 25% (0.37 m/s). Unfortunately, the model found was not really robust against external perturbations.

Videos

Videos of the simulation with different oscillator models (the video goes faster than the real simulation)

Hopf



hopf.avi Model based on Hopf oscillators (without feedback)

Matsuoka



matsuoka.mpeg Model based on Matsuoka oscillators (without feedback)




matsuoka_fb.mpeg Model based on Matsuoka oscillators (with feedback)