Gaze Control Strategy for Vision-Guided Humanoid Walking (Blickwinkelsteuerung für visuell geführtes humanoides Laufen)

Abstract

Abstract To increase the autonomy of a vision guided walking robot, perception must be controlled such that the humanoid is provided with sufficient information about its environment. This article outlines a modular task-specific and situation-dependent gaze control scheme which selects an optimal view direction allowing the goal-oriented acquisition of the relevant information. The basic idea of the approach rests upon maximization of the predicted visual information content of a view situation. For information management, a hybrid Extended Kalman Filter is proposed considering the peculiarities of legged locomotion. Specific view direction evaluation strategies for obstacle avoidance and self-localization tasks are formulated. For dealing with these conflicting objectives a decision strategy for situation-dependent view direction selection based on Utility Theory is presented. Several references to related published work on simulative validation and real-world experiments with different robot platforms are given. The results demonstrate the relevance and practicality of the proposed gaze control approach in the framework of a system for vision-guided humanoid walking.