Techniques for Improving Vision and Locomotion on the Sony AIBO Robot

Abstract

The restricted setting and uniformly prescribed hardware of the Sony Legged League of RoboCup provide an environment for testing algorithms on autonomous robots with a view towards possible applications in real world situations. In this study we show how two techniques - Support Vector Machines and Hill Climbing - can be applied to problems faced by the robots in this league. We use Support Vector Machines to perform collision detection and also to assist the process of color classification, while a Hill Climbing algorithm is employed to improve straight line walking speed through walk parameter optimisation.