analiza aiqbench
The paper proposes a training method for vision-language-action robot models that reweights demonstration and rollout data by how much each action chunk reduces the remaining steps to task success, using reverse search over a goal-reachability graph. It avoids a learned reward model, critic, or progress predictor. The reported 91% success is from four manipulation tasks and three backbones, so the gains need independent replication before they generalize. For ML readers, the useful idea is treating mixed-quality and failed trajectories as recoverable data rather than discarding them.
Graph-Reweighted Behavior Cloning with Human Interventions for Precise Robotic Manipulation
arXiv:2610.20504v1 Announce Type: new Abstract: 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. The key challenge is to identify which action chunks actually advance the task, rather than uniformly cloning or discarding imperfect trajectories. 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. Reverse graph search computes minimal remaining steps to task success within the dataset support, and each action chunk is weighted by how much its execution reduces this potential. 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. New rollouts are incorporated by appending them to the graph and rerunning search, without training a reward model, action-conditioned critic, or progress predictor. 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. More details and videos are available at: https://gr-bc.github.io
Abstract: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. The key challenge is to identify which action chunks actually advance the task, rather than uniformly cloning or discarding imperfect trajectories. 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. Reverse graph search computes minimal remaining steps to task success within the dataset support, and each action chunk is weighted by how much its execution reduces this potential. 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. New rollouts are incorporated by appending them to the graph and rerunning search, without training a reward model, action-conditioned critic, or progress predictor. 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. More details and videos are available at: this https URL
Accepted at CoRL 2026. 25 pages, 16 figures, including appendix. Project page: this https URL
Robotics (cs.RO)
arXiv:2610.20504 [cs.RO]
(or arXiv:2610.20504v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2610.20504 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 300 cuvinte
Subiecte conexe
Hint-based CoT faithfulness evals still mostly work on Claude
LessWrong · 30 iul. 2026
Links #5: 2026/07
LessWrong · 31 iul. 2026
The J-Space Debate, Agent Swarms, and Pacing Frontier AI - Digital Minds Newsletter #4
LessWrong · 18 sept. 2026
How do neural networks regress cubic polynomials? Apparently, they use a trick invented in Milan 500 years ago
LessWrong · 22 iul. 2026