Compliant Residual DAgger: Improving Real-World Contact-Rich Manipulation with Human Corrections

TL;DR

This paper introduces Compliant Residual DAgger, a novel approach that uses human corrections with compliance control to improve contact-rich manipulation tasks, achieving significant success rate improvements with minimal data.

Contribution

The paper presents a new CR-DAgger method with a compliant intervention interface and residual policy learning, enhancing real-world robot manipulation with human corrections.

Findings

Improved success rates by 64% on four tasks.

Effective learning from minimal human correction data.

Outperforms retraining and finetuning approaches.

Abstract

We address key challenges in Dataset Aggregation (DAgger) for real-world contact-rich manipulation: how to collect informative human correction data and how to effectively update policies with this new data. We introduce Compliant Residual DAgger (CR-DAgger), which contains two novel components: 1) a Compliant Intervention Interface that leverages compliance control, allowing humans to provide gentle, accurate delta action corrections without interrupting the ongoing robot policy execution; and 2) a Compliant Residual Policy formulation that learns from human corrections while incorporating force feedback and force control. Our system significantly enhances performance on precise contact-rich manipulation tasks using minimal correction data, improving base policy success rates by 64% on four challenging tasks (book flipping, belt assembly, cable routing, and gear insertion) while outperforming both retraining-from-scratch and finetuning approaches. Through extensive real-world experiments, we provide practical guidance for implementing effective DAgger in real-world robot learning tasks. Result videos are available at: https://compliant-residual-dagger.github.io