Imitation Learning from Pixel Observations for Continuous ControlDownload PDF

Published: 28 Jan 2022, Last Modified: 13 Feb 2023ICLR 2022 SubmittedReaders: Everyone
Keywords: imitation learning, optimal transport, GAIL, adversarial learning
Abstract: We study imitation learning using only visual observations for controlling dynamical systems with continuous states and actions. This setting is attractive due to the large amount of video data available from which agents could learn from. However, it is challenging due to $i)$ not observing the actions and $ii)$ the high-dimensional visual space. In this setting, we explore recipes for imitation learning based on adversarial learning and optimal transport. A key feature of our methods is to use representations from the RL encoder to compute imitation rewards. These recipes enable us to scale these methods to attain expert-level performance on visual continuous control tasks in the DeepMind control suite. We investigate the tradeoffs of these approaches and present a comprehensive evaluation of the key design choices. To encourage reproducible research in this area, we provide an easy-to-use implementation for benchmarking visual imitation learning, including our methods.
One-sentence Summary: We propose strong recipes for imitation learning from visual observations only based on adversarial learning and optimal transport.
14 Replies