Augmenting movement imitation with reflexive stability behavior

Even though humans and a humanoid robots have similar kinematic structures, directly imitating human movement does not guarantee the same behavior of the robot because of their different dynamical properties and perhaps different conditions under which the task is being performed. Our research starts by showing how to apply a generalization algorithm to extract the desired movement primitives from multiple human demonstrations of the same movement. The emphasis of the paper is on a method that constrains the extracted movement primitives when mapping them to a robot, taking into account a critical criterion of the task. As a practical example we study the stability of a robot, which is determined through a normalized zero-moment-point. Our approach is based on prioritized task control and allows direct movement transfer as long as the selected criterion is met. It only constrains the movement when the criterion approaches a critical condition. The critical condition thus triggers a reflexive, subconscious behavior, which has higher priority than the desired, conscious movement. Augmenting movement imitation with the proposed approach allows direct movement imitation, which can be executed on the robot at different ground inclination conditions. We demonstrate the properties of the algorithm on a real, human-inspired leg robot developed in our laboratory.