Visuomotor Control

We train the action sequence of the robotic hand by watching the target image in this experiment, where the initial guess is a still hand.

All algorithms try to optimize a visually similar result, however, some fail because of occlusions and aliases. Only ours reconstruct the correct motion.