The quadruped ALoF and a step towards real world haptic terrain classification

摘要

The intention of this paper is twofold. It first describes the robotic platform AloF and its control software framework which was designed for autonomous locomotion in rough terrain. The robot has a very large range of leg motion to actively explore its surroundings through haptic interaction and to increase its locomotion capabilities. The platform is robust enough to carry adequate sensors to perceive the surrounding environment. As a step towards reliable terrain classification for legged robots, we are additionally presenting a novel method based on haptic feedback. The method has been evaluated on a simplified ALoF leg, employed in a test setup. Eleven different samples of natural terrains have been classified in experiments. Features, extracted primarily from contact force and motor current measurements, were used for training and prediction by a multiclass AdaBoost machine learning algorithm. About 87% of the real terrain samples have been classified correctly.