Sensor-based failure prediction in biped locomotion

In order for robots being able to safely move among humans, they have to be able to adapt their motor behavior to unexpected situations. In this paper, we are interested in achieving a predictive model of sensor traces that enables to detect failures in advance. This failure system associates experienced sensor information with corresponding robot skills. Specifically, we propose a bio-inspired architecture that adapts online the generated biped locomotion using sensory feedback. The proposed failure detection system of the biped locomotion is implemented based on Associative Skill Memories (ASM), extending the concept to the specific case of biped locomotion. Promising results were achieved, with failures being detected on average 2.23 seconds in advance.