Biorobotics Laboratory BioRob

Control of Locomotion in Modular Robotics

This master project focuses on the control of locomotion in modular robotics. We are particularly interested in applying the Central Pattern Generator (CPG) approach to the modular robot YaMoR. The concept of CPG has been introduced in the eighties to explain the mechanism of locomotion in vertebrates. A CPG allows to control multiple antagonist muscles and to modulate the generated pattern with simple high-level stimuli.

In YaMoR, the CPG is modelized as as system of oscillators, whose outputs control the servo motors of the modules. These oscillators contain many free control parameters that have to be tuned for creating a satisfying pattern on the whole mounted robot. As the number of parameters increases, it becomes advantageous to use learning algorithms.

It was shown previously (Online optimization of modular robot locomotion) that an algorithm, called Powell's method, gives outstanding results in simulations. Powell's method is a fast and simple heuristic, generally used on mathematical functions, for finding the minimum of a multi-dimensional function.

This project mainly aims at validating the results of the simulations on the real robot platform YaMoR. To prove the efficieny of Powell's method, a comparison with another algorithm is also proposed.