Crawling infants and dynamical systems: mathematical models and 
applications to legged locomotion


Rhythmic movements in animals, and especially locomotion, are controlled 
mainly by Central Pattern Generators (CPGs). These neural networks are 
responsible for the generation of complex rhythmic patterns and are 
often modeled as coupled oscillators. CPGs are becoming a very popular 
tool for the control of locomotion of legged robots, since their main 
advantages are their stability to perturbations (limit cycle behavior) 
and their synchronization capabilities that allow strong coupling with 
the robot and the environment.

In this presentation, I will show our work done on CPG controllers 
during the RobotCUB project, which goal is to build a 2-year-old 
infant-like humanoid robot able to interact with its environment. First 
experimental results on infants crawling compared to other quadrupeds 
will be presented. Then i will show how mathematical models of CPGs can 
reproduce features of quadruped locomotion. It includes new, flexible, 
oscillator models, generic design of networks of coupled dynamical 
systems and automatic construction of limit cycles of arbitrary shape 
(with learning embedded in the dynamical system). Finally, results on 
several robots (from very stiff robots to ones with passive dynamics) 
will demonstrate how our CPGs adapt to the dynamics of the robots and 
emphasize the importance of sensory feedback during locomotion.