PGDG: Physically Grounded Data Generation for Robust Bimanual Policy Learning from a Single Demonstration

TL;DR

PGDG is a novel data generation framework that enhances single demonstration behavior cloning for bimanual manipulation by creating diverse, physically plausible recovery behaviors, improving success rates in simulation and real-world tasks.

Contribution

The paper introduces PGDG, a zero-shot data curation method that expands a single demonstration into a rich dataset of recovery behaviors without extra labeling.

Findings

Success rate on RotateBox-Pitch increased from 38% to 93% in simulation.

Success rate on RotateBox-Pitch increased from 35% to 82% in the real world.

Fine-tuning foundation models like GR00T improved success from 46% to 77%.

Abstract

Behavior cloning for contact-rich bimanual manipulation remains challenging because diverse demonstrations are expensive to collect, and even small disturbances can push the system into off-manifold states where no recovery supervision is available. We propose PGDG, a data generation framework with zero-shot curation that expands a single demonstration into a compact dataset of physically plausible, successful, and diverse recovery behaviors without additional human labeling. PGDG iterates between a physics-grounded sampler and a dataset curator, where the curator selects informative, non-redundant, and recoverable behaviors to update the sampling distribution toward under-covered recovery modes, and the sampler draws physically plausible rollout candidates from this updated distribution and retains successful trajectories. To further improve data quality, PGDG applies short-horizon sampling-based control to relabel selected risky states with corrective actions. Across four bimanual manipulation tasks, PGDG consistently outperforms spatial-only augmentation in both simulation and zero-shot real-world transfer. On RotateBox-Pitch, success improves from 38% to 93% in simulation and from 35% to 82% in the real world. PGDG also enables effective foundation models fine-tuning such as GR00T, increasing success from 46% to 77%. Additional results are available in our website: https://cunxid.github.io/PGDG/.