Imitation Learning

We reconstruct the action sequence of the robotic hand by watching a target video in this experiment, whose values are randomly initialized.

Only ours reconstructs the whole sequence without biases or magnitude shifts.