Compliant residual DAgger: Improving real-world contact-rich manipulation with human corrections
Compliant residual DAgger: Improving real-world contact-rich manipulation with human corrections
This research addresses challenges in Dataset Aggregation (DAgger) for contact-rich manipulation by introducing Compliant Residual DAgger (CR-DAgger). The system features a compliant intervention interface for gentle human corrections without interrupting robot execution and a residual policy formulation that incorporates force feedback. Experiments show it improves base policy success rates by 64% on four tasks (book flipping, belt assembly, cable routing, gear insertion) using minimal correction data, outperforming retraining-from-scratch and fine-tuning approaches.
Full article
Article text is not available for this item. Read it at the original source.
Read full article at sourcedoi.org · Opens in a new tab; source language may differ.
The paper introduces Compliant Residual DAgger (CR-DAgger), a method combining compliant intervention interfaces and residual policy formulations to improve contact-rich manipulation tasks by integrating human corrections with force feedback, achieving a 64% improvement in success rates on four specific benchmarks.
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
Reported success rate
64%
View original evidence
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, b
Open source S3
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Environment, Control, Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
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, b
Open source S3
elt 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/ .
Open source S4
Source：International Journal of Robotics Research — Journal metadata · doi.org