aiqbench analysis
This research addresses the computational inefficiency of training independent models for every robot manipulation task by proposing a knowledge distillation framework based on Conditional Flow Matching. By transferring learned velocity fields from single-task experts to a shared policy, the method improves multi-task performance without increasing model capacity or sacrificing fidelity to expert demonstrations. This advancement highlights a significant step toward scalable generative modeling in robotics, where efficiency and adaptability are critical for real-world deployment.
Distillation for Efficient Multitask Manipulation Policies via Conditional Flow Matching
arXiv:2609.28107v1 Announce Type: cross Abstract: Advances in generative modeling have recently been extensively employed in robotics for policy learning. In particular, Conditional Flow Matching (CFM) trained with expert demonstrations has been shown to outperform existing methods on robot manipulation benchmarks. While prior work has mainly focused on single-task settings, we study the problem from a multi-task perspective, as training independent models for each task is computationally expensive. Multi-Task policy learning comes with its own set of challenges, as naively training on a concatenated dataset of demonstrations would either require increased model capacity to accommodate the added complexity or result in drops in performance. We propose to distill knowledge from single-task CFM experts into a shared multi-task policy by transferring their learned velocity fields. We combine this distillation signal with the original CFM objective to retain fidelity to the demonstrations. Experiments on RLBench show that our approach improves multi-task policy performance over naive training while maintaining a fixed model size.
Abstract:Advances in generative modeling have recently been extensively employed in robotics for policy learning. In particular, Conditional Flow Matching (CFM) trained with expert demonstrations has been shown to outperform existing methods on robot manipulation benchmarks. While prior work has mainly focused on single-task settings, we study the problem from a multi-task perspective, as training independent models for each task is computationally expensive. Multi-Task policy learning comes with its own set of challenges, as naively training on a concatenated dataset of demonstrations would either require increased model capacity to accommodate the added complexity or result in drops in performance. We propose to distill knowledge from single-task CFM experts into a shared multi-task policy by transferring their learned velocity fields. We combine this distillation signal with the original CFM objective to retain fidelity to the demonstrations. Experiments on RLBench show that our approach improves multi-task policy performance over naive training while maintaining a fixed model size.
Robotics (cs.RO); Artificial Intelligence (cs.AI); Machine Learning (cs.LG)
arXiv:2609.28107 [cs.RO]
(or arXiv:2609.28107v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.28107 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.
Full text under the publisher's open licence · 253 words
Related coverage
XS-VLA: Coupling Coarse-grained Spatial Distillation with Latent Flow Matching for Lightweight Robotic Control
arXiv · Jul 28, 2026
MINERVA: How Small Can a Manipulation Policy Be and Still Solve LIBERO?
arXiv · Sep 4, 2026
Implementing and Evaluating a Basic Per-Action Monitor for Safer Evals
METR · Sep 27, 2026
Runway’s WorldPrompt and the Engineering of Real-Time Worlds
Latent Space · Sep 25, 2026