Dynamically Stepping Over Obstacles by the Humanoid Robot HRP-2

Abstract

Humanoid robots have the potential to navigate through complex environments such as a standard human living environment. One of the advantages is that a biped can negotiate obstacles by stepping over them, which is the topic of the work presented in this paper. The main focus of this research is to investigate stepping over large obstacles. Previous work has reported on algorithms using quasi-static balancing, which resulted in somehow unnatural slow motions. This work however, focuses on stepping over larger obstacles in a fluent dynamic motion, using stability criteria on zero moment point instead of center of gravity. All the work is formulated in function of the elaborate HRP-2 humanoid research platform. The strategy uses a preview controller for dynamic balancing and consists of collision free trajectory generation for the feet and waist. The paper provides a proof of concept for dynamically stepping over large obstacles by experiment and pinpoints some difficulties in the planning for practical implementations. Several measures have been implemented e.g., to avoid overstretching of the knee and reduce impact at touch-down