Negotiating Large Obstacles with a Humanoid Robot via Multi-Contact Motion Planning

Incremental progress in humanoid robot locomotion over the years has achieved essential capabilities such as navigation over at or uneven terrain, stepping over small obstacles and imbing stairls. However, the locomotion research has mostly been limited to using only bipedal gait and only foot contacts with the environment, using the upper body for balancing without considering additional external contacts. As a result, challenging locomotion tasks like climbing over large obstacles relative to the size of the robot have remained unsolved. In this paper, we address this class of open problems with an approach based on multi-contact motion planning, guided by physical human demonstrations. Our goal is to make humanoid locomotion problem more tractable by taking advantage of objects in the surrounding environment instead of avoiding them. We propose a multi-contact motion planning algorithm for humanoid robot locomotion which exploits the multi-contacts at the upper and lower body limbs. We propose a contact stability measure, which simplies the contact search from demonstration and contact transition motion generation for the multi-contact motion planning algorithm. The algorithm uses the whole-body motions generated via Quadratic Programming (QP) based solver methods. The multi-contact motion planning algorithm is applied for a challenging task of climbing over a relatively larger obstacle compared to the robot. We validate our planning approach with simulations and experiments for climbing over a large wooden obstacle with COMAN, which is a complaint humanoid robot with 23 degrees of freedom (DOF). We also propose a generalization method, the \Policy-Contraction Learning Method" to extend the algorithm for generating new multi-contact plans for our multi-contact motion planner, that can adapt to changes in the environment. The method learns a general policy and the multi-contact behavior from the human demonstrations, for generating new multi-contact plans for the obstacle-negotiation.