PREDICTIVE REACHING CONTROL WITH MULTIPLE MOTION MODELS

: In this work we address the problem of controlling the arm of a humanoid robot to reach for moving objects. 3D target’s trajectory is measured by the robot’s active stereo head with color based segmentation and tracking methods. Future positions of the target are predicted, at an appropriate time horizon, by fusing the information from multiple motion model estimators, including constant velocity, acceleration, circular and periodic motions. The arm positioning system is controlled by setting its reference to the target’s position at the prediction horizon, to cope with the arm slow dynamics. Experimental results show that, compared to a non-predictive approach, the proposed method reduces the average tracking error in about 50%.