What Matters in Language Conditioned Robotic Imitation Learning over Unstructured Data

TL;DR

This paper investigates critical challenges in language-conditioned robotic imitation learning from offline data, proposing architectural improvements and a novel model that significantly outperforms existing methods on complex manipulation benchmarks.

Contribution

It introduces a comprehensive analysis of design choices and presents a new model with hierarchical control, multimodal transformers, and contrastive learning, advancing the state of the art.

Findings

Significant performance improvements on CALVIN benchmark

Effective use of hierarchical decomposition and multimodal transformers

Open-sourced code and models for future research

Abstract

A long-standing goal in robotics is to build robots that can perform a wide range of daily tasks from perceptions obtained with their onboard sensors and specified only via natural language. While recently substantial advances have been achieved in language-driven robotics by leveraging end-to-end learning from pixels, there is no clear and well-understood process for making various design choices due to the underlying variation in setups. In this paper, we conduct an extensive study of the most critical challenges in learning language conditioned policies from offline free-form imitation datasets. We further identify architectural and algorithmic techniques that improve performance, such as a hierarchical decomposition of the robot control learning, a multimodal transformer encoder, discrete latent plans and a self-supervised contrastive loss that aligns video and language representations. By combining the results of our investigation with our improved model components, we are able to present a novel approach that significantly outperforms the state of the art on the challenging language conditioned long-horizon robot manipulation CALVIN benchmark. We have open-sourced our implementation to facilitate future research in learning to perform many complex manipulation skills in a row specified with natural language. Codebase and trained models available at http://hulc.cs.uni-freiburg.de