Left to right, top to button. The original actions, the mirrored actions, the noisy actions (randomly added Gaussian noise into the original actions), mirrored action with Gaussian noise.
Left to right. The source actions (training), target actions (training), and the generated actions