Toward automatic robot programming: learning human skill fromvisual data

摘要

We propose a novel approach to program a robot by demonstratingnthe task multiple number of times in front of a binocular vision system.nWe track artificially-induced features appearing in the image plane duento nonimpedimental color stickers attached at different fingertips andnwrist joint, in a simultaneous feature detection and tracking framework.nA Kalman filter does the tracking by recursively predicting thententative feature location and a higher order statistics (HOS)-basedndata clustering algorithm extracts the feature. A fast and efficientnalgorithm for the vision system thus developed processes a binocularnvideo sequence to obtain the trajectories and the orientationninformation of the end effector from the images of a human hand. Thenconcept of trajectory bundle is introduced to avoid singularities and tonobtain an optimal path