Self-Supervised Disentangled Representation Learning for Third-Person Imitation Learning
This addresses the challenge of expensive first-person demonstration collection for robots by enabling imitation from third-person videos, though it is incremental as it builds on existing TPIL concepts with a novel method for a specific bottleneck.
The paper tackles the problem of third-person imitation learning (TPIL) for robot tasks with egomotion, where first-person and third-person views differ visually, by proposing a disentangled representation learning method that uses a dual auto-encoder with specific losses, resulting in demonstrated effectiveness in experiments.
Humans learn to imitate by observing others. However, robot imitation learning generally requires expert demonstrations in the first-person view (FPV). Collecting such FPV videos for every robot could be very expensive. Third-person imitation learning (TPIL) is the concept of learning action policies by observing other agents in a third-person view (TPV), similar to what humans do. This ultimately allows utilizing human and robot demonstration videos in TPV from many different data sources, for the policy learning. In this paper, we present a TPIL approach for robot tasks with egomotion. Although many robot tasks with ground/aerial mobility often involve actions with camera egomotion, study on TPIL for such tasks has been limited. Here, FPV and TPV observations are visually very different; FPV shows egomotion while the agent appearance is only observable in TPV. To enable better state learning for TPIL, we propose our disentangled representation learning method. We use a dual auto-encoder structure plus representation permutation loss and time-contrastive loss to ensure the state and viewpoint representations are well disentangled. Our experiments show the effectiveness of our approach.