Generation of Complex Dynamic Motion for a Humanoid Robot

Whole-body coordinated motion of a humanoid robot is known to be a difficult problem for the robotics community. Factors like the dynamic model of the robot, its physical limitations, and its stability throughout the motion must be considered for the results to be implementable on a real robot. Redundancy poses great difficulties as the control of a certain part of the humanoid, if not well planned, can produce non desirable effects on other parts, generating motions that look by far unnatural as compared to the way a human being would do it. This thesis has as starting point an inverse dynamics control scheme based on quadratic programming optimization. This scheme was extended with different tasks to control the different parts of the robot body satisfying at the same time the dynamic stability condition and the joint limits constraints. The implemented dynamically controlled visual task makes the motion of the head more realistic. The results of the control scheme consist on the robot HRP-2 sitting down on a chair. To this end, there is control over the head, chest, waist, both hands, both feet and the grippers, at different times, but often coinciding with more than one prioritized task at the same time. The rigid contacts are not limited to the ground surface with the feet, but extended to the hands with the armrest. These additional contacts make it possible for the robot to sit down in a realistic way. Human imitation is another complex area for whole-body motion. The second result presented in this thesis involves the imitation of a person’s dance by the robot. The motion was first acquired from a dancer using a motion capture system available at the laboratory, and then the robot’s joints were recovered using a forward kinematics optimization scheme. The motion thus obtained was further dynamically processed using the inverse dynamics control scheme to generate a dynamically stable motion for the robot since the ZMP condition is considered in the model. The motion is additionally modified using arbitrary tasks on the operational space or directly on the joint space in order to make the robot imitate the original motion more closely or even purposely differ from it at some point. The availability of these fast and easily realizable modifications, while maintaining dynamic stability, gives special advantages to the method proposed for imitation.