Self-Supervised Imitation Learning

Pierre Sermanet1*, Corey Lynch1*†, Yevgen Chebotar2*, Kelvin Xu1†, Jasmine Hsu1, Eric Jang1, Stefan Schaal2, Sergey Levine1
1 Google Brain, 2 University of Southern California
(* equal contribution, † Google Brain Residency program


The general goal of this project is to give to robots the ability to learn to imitate humans from observation and without labels.
We present the following papers as steps toward that goal.

1. Time-Contrastive Networks (TCN)

2. Unsupervised Perceptual Rewards

3. TCN update: Learning to imitate