Human-Inspired Anticipative Postural Control Architecture for Humanoid Robots

Abstract

The study of Human Postural Control (HPC) system reveals the existence of an anticipative component for improving its performance. This anticipative subsystem is able to preview future consequences caused by a postural disturbance and, as well, to trigger corrective actions. This paper presents the novel high level architecture for the humanoid robot TEO based on the same principles than the human system. The core of the architecture is an anticipative system for predicting postural corrections based on the evaluation of unexpected events or surprises. Basically, these events are caused by the mismatch between sensory perceptions and the expected. The resulting surprise is used by a decision making system that decides whether a correction is necessary, and selects the adequate graded postural correction from a set of motion patterns or synergies. All the anticipative process is enabled by Neuro-Fuzzy modules that provide the system with human style reasoning behaviour.