Real-time stabilization of a falling humanoid robot using hand contact: An optimal control approach

Abstract

This paper introduces a real-time strategy to stabilize a falling humanoid robot by making hand contact with rails or walls in the environment. It uses an optimal control strategy with a simplified three-link model and finds an optimal hand contact using a direct shooting method. The objective function is designed to maintain stability while minimizing the probability of the contact slipping and minimizing impact that may damage the robot's arm. To achieve real-time performance, the method uses a precomputed database of necessary sticking friction coefficients at the contact points for all possible postimpact states. Validation is performed on a number of simulated falls in several rails and walls.