Evolution of neural controllers for locomotion and obstacle avoidance in asix-legged robot

This article describes how the SGOCE paradigm has been used within the cont ext of a 'minimal simulation' strategy to evolve neural networks controllin g locomotion and obstacle avoidance in a six-legged robot. A standard genet ic algorithm has been used to evolve developmental programs according to wh ich recurrent networks of leaky-integrator neurons were grown in a user-pro vided developmental substrate and were connected to the robot:sensors and a ctuators. Specific grammars have been used to limit the complexity of the d evelopmental programs and of the corresponding neural controllers. Such con trollers were first evolved through simulation and then successfully downlo aded on the real robot.