Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation

Key points

• Long-horizon precise manipulation produces mixed-quality data for post-training vision-language-action (VLA) models: even successful demonstrations and intervention-assisted rollouts may contain detours, hesitations, failed attempts, and imperfect recoveries.1
• The key challenge is to identify which action chunks actually advance the task, rather than uniformly cloning or discarding imperfect trajectories.1
• This consequence-based weighting reuses recoverable segments from failed trajectories, leverages interventions as recovery paths, and down-weights locally regressive chunks even inside successful trajectories.1
• Across four precise manipulation tasks and three VLA backbones, GR-BC improves single-round supervised fine-tuning and reaches 91% success after a few offline iterations, without trajectory comparisons or progress labels.1

Sources (1)

• [1]Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 14, 05:09 PM

  We propose Graph-Reweighted Behavior Cloning (GR-BC), an offline iterative post-training framework that builds a task-conditioned goal-reachability graph over demonstrations and iteratively collected rollouts, including both successful and failed episodes with human interventions.

  Long-horizon precise manipulation produces mixed-quality data for post-training vision-language-action (VLA) models: even successful demonstrations and intervention-assisted rollouts may contain detours, hesitations, failed attempts, and imperfect recoveries.

Extractive summary: sentences quoted from the sources.