From lampreys to salamanders evolving neural controllers for swimming and walking

This paper presents how neural swimming controllers for a lamprey can be adapted for controlling both the swimming and the walking of a salamander like an imat Using a Genetic Algorithm GA we extend a connectionist model of the biological Central Pat tern Generator CPG controlling the swimming of a lamprey Ekeberg to control the locomotion of a D mechanical simulation of a salamander We rst summarize experiments on the evolution of alternative swimming controllers for the lamprey Ijspeert et al The aim of that work was to study whether there exists other neural con gurations than that found in the lamprey which could control swimming with the same e ciency and to develop a method for developing neural locomotion control lers using a GA We then present how that method namely a staged evolution of the neural con gura tion of a connectionist model can be used to extend swimming controllers to control both swimming and walking Controllers which similarly to CPGs of anim als can produce complex oscillations when receiving simple excitatory signals are thus developed In par ticular we generate a controller which can switch from swimming to walking and produce di erent speeds of motion depending on its excitation