Visuomotor Control

We train the action sequence of the rod by watching the target image in this experiment, whose values are randomly initialized.

Only ours-no-grad and ours converge to a reasonable action sequence leading to the given target image.