Self-protect falling trajectories for humanoids with resilient trunk

Efficient humanoid robots must have the ability to protect themselves during unexpected falls. In this paper, a self-protect falling trajectory library for humanoid robots is proposed by means of a 4-link model and to be triggered and used on-line, by the controller, in the course of fallings that do not result in a total loss of contacts. When possible, the trajectory is generated considering intermediary contact with the ground, which directly affects the motion state of the system. A collision model is added to the optimization process to fully describe the fall process. Off-line recorded instructed human fall motions is used as an initial guess for the trajectory optimization process. Then, we use B-Spline to match optimization curves and store them as control points so the robot can quickly select and switch a trajectory that matches the current robot state. Therefore, the robot can cope with all the fall directions and different magnitudes of impact considered in the library. Finally, the simulations and real experiments of three directions (forward, backward, and lateral) of falling protection are performed to assess the effectiveness of our proposed method.