Efficient self-collision avoidance based on focus of interest for humanoid robots

Abstract

This paper deals with the self-collision avoidance problem for humanoid robots in an efficient way. Self-collision avoidance is introduced as a constraint for each task in a hierarchical Inverse Kinematic (IK) problem. Since the number of link pairs which needs to be updated and checked for self-collision, in every control loop, is large, the novel concept of Self-Collision Avoidance Focus of Interest (SCAFoI) is proposed. SCAFoIs permits to predict and dynamically select the necessary link pairs to be checked online to improve the computation efficiency. For each of the several SCAFoIs, which corresponds to the related pairs of kinematic chains of the whole body, the status of the relative positional relationship is predicted. The prediction is done using a Support Vector Machine (SVM) which is a widely used classifier from the machine learning field. Moreover, techniques are proposed to guarantee and improve the prediction performance of the trained classifier. The effectiveness of the framework is verified using the whole-body motion control library OpenSoT by simulation on the model of the recently developed humanoid robot WALK-MAN.