Applying genetic programming to evolve behavior primitives and arbitrators for mobile robots

Abstract

The behavior-based approach has been successfully applied to designing robot control systems. This paper presents our work, based on evolutionary algorithms, to program behavior-based robots automatically. Instead of hand-coding all the behavior controllers or evolving an entire control system for an overall task, we suggest our approach at the intermediate level: it includes evolving behavior primitives and behavior arbitrators for a mobile robot to achieve the specified tasks. To examine the developed approach, we evolve a control system for a moderately complicated box-pushing task as an example. We first evolved the controllers in a simulation and then transferred them to the Khepera miniature robot. Experimental results show the promise of our approach, and the evolved controllers are transferred to the real robot without loss of performance.