SEIL: Simulation-augmented Equivariant Imitation Learning

TL;DR

SEIL enhances imitation learning in robotic manipulation by combining simulation-based data augmentation with symmetry-aware models, enabling efficient learning from minimal demonstrations.

Contribution

It introduces a novel simulation-augmented data augmentation strategy combined with an equivariant model exploiting symmetry, improving sample efficiency in imitation learning.

Findings

Learns complex manipulation tasks with fewer than ten demonstrations

Outperforms baseline methods significantly in experiments

Exploits $ ext{O}(2)$ symmetry for improved learning efficiency

Abstract

In robotic manipulation, acquiring samples is extremely expensive because it often requires interacting with the real world. Traditional image-level data augmentation has shown the potential to improve sample efficiency in various machine learning tasks. However, image-level data augmentation is insufficient for an imitation learning agent to learn good manipulation policies in a reasonable amount of demonstrations. We propose Simulation-augmented Equivariant Imitation Learning (SEIL), a method that combines a novel data augmentation strategy of supplementing expert trajectories with simulated transitions and an equivariant model that exploits the $\mathrm{O}(2)$ symmetry in robotic manipulation. Experimental evaluations demonstrate that our method can learn non-trivial manipulation tasks within ten demonstrations and outperforms the baselines with a significant margin.