No abstract

This paper addresses the problem of automotie akill acquisition by a mbot. It reports that six triols of a reach-grasp-release-rYt~ct skill are suflicient for learning a canonical description of the task under the follouting circumstances: The mbot is Robonaut, NASA's space-capable, desterous humanoid. Robonaut was teleoperated by a person using full immersion Virtual Reality technology that tmnsforms the openator's a n and hand motions into those of the robot. The operator's sole S O U I C~ of real-time feedback was visual. DUTing the s*: trials all of the Robot's senaory inputs and motor control pammeters were recorded as time-series. Later the time-series from each trial was paditioned into the same number of episodes as a function of changes in the motor parameter sequence. The episodes were time normalized and avemged D C M S I trials The resultant motor parameter sequence and sensor signals w e r~ used to eontml the mbot without the teleopemtor. The mbot wuas able to perform the task autonomously with mbot starting positions and object locations both similar to, and different from the original trials.Fig. 1. Robonaut, NASA's space capable humanoid robot

No abstract

No abstract

No abstract