Robbie the running robot: a distributed learning system

Abstract

A unique distributed control system has been developed to allow a hexapod robot to learn how to walk. It models competing directives which drive the robot to walk as quickly as possible without falling over given minimal initial knowledge. The system architecture features highly decentralised arrangement of cooperating modules with no central controller. This has many advantages over conventional control systems, including robustness, scope for incremental development and the ability to facilitate accurate pre-prototype simulation. Accelerated learning was achieved by using the robot's intrinsic knowledge of its own symmetry to infer additional information about its surroundings. This was found to improve the robot's performance, particularly at the start of the learning process when information was limited. These concepts can be applied to autonomous robots for use in deep sea and other inhospitable environments.