Biomechanically motivated lateral biped balancing using momentum control

A humanoid robot is expected to experience various force disturbances during interaction with humans and when accidentally bumping into objects. For the safety of the surrounding humans and the robot itself, the robot should prevent a fall. In some situations when a step cannot be taken, rapid body movement should be generated to maintain balance. We will show that for robots this often occurs after a lateral push and therefore deserves special attention. First the biomechanics of the human reflex motion are analyzed through experiments with different push forces. The motion capture system gives the position of all markers placed on the human body, which are converted to robot joint angles by an algorithm of inverse kinematics using optimization techniques. This approach handles differences in degrees of freedom and additional user-defined goals in a natural way. The motions are accurately imitated by the robot, and although they are less useful for withstanding actual forces, the algorithm can be used for other applications and gives insight in the robot’s limiting factors to perform human motion. Second, these human strategies are explained through the dynamics of simple and intuitive 2D models of a humanoid. A disturbance force is added to the existing Cart-Table model and angular momentum is taken into account. These models are then extended to the general 3D case. It is shown that stability can be maintained and the Zero-Moment Point (ZMP) and Center of Mass (CoM) can be regulated by controlling the rate of momentum, which in turn can be achieved by controlling the joint accelerations. The relations are derived for both an intuitive 2D and general 3D case. The proposed controller can even handle the case when the biped is in partial contact with the ground, is verified in simulation, and is a promising feedforward setup for balancing.