Learning Control by Iterative Inversion
Gal Leibovich1*, Guy Jacob1*, Or Avner2*, Gal Novik1, Aviv Tamar2
1 Intel Labs
2 Technion - Israel Institute of Technology
* equal contribution, order determined by coin toss
Gal Leibovich1*, Guy Jacob1*, Or Avner2*, Gal Novik1, Aviv Tamar2
1 Intel Labs
2 Technion - Israel Institute of Technology
* equal contribution, order determined by coin toss
We propose iterative inversion -- an algorithm for learning an inverse function without input-output pairs, but only with samples from the desired output distribution and access to the forward function. The key challenge is a distribution shift between the desired outputs and the outputs of an initial random guess, and we prove that iterative inversion can steer the learning correctly, under rather strict conditions on the function.
We apply iterative inversion to learn control. Our input is a set of demonstrations of desired behavior, given as video embeddings of trajectories (without actions), and our method iteratively learns to imitate trajectories generated by the current policy, perturbed by random exploration noise. Our approach does not require rewards, and only employs supervised learning, which can be easily scaled to use state-of-the-art trajectory embedding techniques and policy representations. Indeed, with a VQ-VAE embedding, and a transformer-based policy, we demonstrate non-trivial continuous control on several tasks. Further, we report an improved performance on imitating diverse behaviors compared to reward based methods.
Note: None of the reconstruction videos below were cherry-picked.
5 samples; 64 frames
5 samples; 64 frames
Sample #1; 64 frames
Sample #2; 64 frames
Sample #1; 128 frames
Sample #2; 128 frames
Sample #3; 128 frames