Biorobotics Laboratory BioRob


Roombots - Central Pattern Generators, 
Symmetries and online learning.

Alexandre Tuleu
January 13, 2009



 This semester pro ject focus on the problem of locomotion in Modular Robotics. This project is a continuation of the work of several former Biologically Inspired Robotic Group (BIRG) students on this topic.

Since modular robots are made of several identical modules, that can assemble each other in almost an infinite numbers of configurations, these works give priority applying a Central Pattern Generator (CPG) approach, rather than a more classical one, where a fine model of the robot kinematics and dynamics is needed. 

The Central Pattern Generators models are well known in Biologicaly field, to be responsible of the generation of rythmics mouvement in vertebrate. Most of them can modulate the pattern created just by acting on a few parameters, and therefore, they are powerful tool to encode the moves. Former works has lead to methods were, given a certain modular robot, we manually design a CPG for it, and then run online optimization algorithm, in order to make the robot learn an optimal gait. 

Most of the former work was made on the YaMoR modular plateform, and have to be updated on a new one, currently develloped at BIRG : the Roombots modules. Indeed, this module add new possibility, as self-reconfiguration, and therefore we want the Roombots robots to be able to automatically determine a CPG. Making this task to still be computed online is a challenging problem. 
In this document, we will first describe some software tools we made for manipulating Central Pattern Generators, Optimization Algorithms, and make the Roombots simulation more accurate, by enhancing the collision detection in 

the Webots simulation software. Then we present a study of the locomotion of a specific Roombots robots, and then we propose an algorithm for automatically design CPG models for modular robot.

Report and presentation


Videos : Demonstration of the collision detection algorithm created for Roombots. : A gait were parameters are randomly choose every 15 seconds. Direction of the gait is inversed by just changing the phase difference of the thorax (central) joint. : Optimal gait found by Particle Swarm Optimization, with a fixed frequency of the CPG of 0.25 Hz. : Optimal gait found by Particle Swarm Optimization, with the frequency being optimized between 0 and 1 Hz. Optimal frequency is not the maximal one, but arround 0.81 Hz.

Code documentation

 You can found on these pages the documentation of the code made in Doxygen format.