Study on falling backward of humanoid robot based on dynamic multi objective optimization

Abstract

Humanoid robots generally have the problem of losing balance and falling down when voltage of joints decreases, which results in damage of hardware devices like mechanical structure and circuit control board. In order to deal with this problem, we simplified the stable control model of a humanoid robot into a quadruple inverted pendulum, corresponding to the robot???s shanks, thighs, torso, and arms, and regarded the bipedal locomotion in the horizontal direction as the movement of cart foundation of the inverted pendulum. We analyzed differential equations of motion and conditions of the complete stability of the quadruple inverted pendulum, imitated and performed falling down of human with protective postures. Combining with kinematics and physical constraints, we implemented a dynamic multi objective optimization algorithm to optimize the angles and angular speed of each joints. This helped the deceleration of the robot???s falling down process and reduced the ground impact force that achieved minimum momentum consumption and immediate stabilization after robot falling down, which can mitigate the damage of the robot???s hardware devices. We did simulation experiments and verified the effectiveness of the method.