Disney research engineers have made it possible for a robot to learn policies from unstructured motion data. First, they extracted a latent space encoding by training a variational autoencoder and then taking short windows of motion from unstructured data as input.
Next, they used the embedding from the time-varying latent code to train a conditional policy in a second stage, thus providing a mapping from kinematic input to dynamics-aware output. By keeping these two stages separate, the team was able to benefit from self-supervised methods to get better latent codes and explicit imitation rewards to avoid mode collapse. The efficiency and robustness was demonstrated in simulation, with unseen user-specified motions, and on a bipedal robot, where dynamic motions were brought to the real world.
- LEGO Ideas Disney Hocus Pocus building set (21341) – Relive hilarious scenes from the classic Disney Hocus Pocus Halloween movie with this detailed...
- 6 LEGO minifigures – Winifred Sanderson, Sarah Sanderson, Mary Sanderson, Max Dennison, Dani Dennison and Allison Watts, plus a Thackery Binx (in...
- Fun features – Spin the water wheel to make pink ’smoke’ puff from the chimney, activate the LEGO light brick to light the fire under the...
