aiqbench analysis
TraceFlow advances robot learning by leveraging a frozen flow-matching policy that incorporates both successful and failed rollout traces without requiring weight updates. This method effectively uses terminal outcome bits from the robot's own failures as negative evidence to correct action densities, significantly improving task success rates in real-world packing and stacking scenarios. The approach demonstrates how test-time adaptation can enhance VLA models using minimal labeled data while maintaining stability across diverse robotic benchmarks.
TraceFlow: Guiding Frozen Flow-Matching Robot Policies with Success and Failure Traces
arXiv:2609.20646v1 Announce Type: new Abstract: A vision-language-action (VLA) policy with a flow-matching action expert generates each action chunk (a short command sequence) by integrating a learned velocity field; once its weights are fixed, the success or failure of an earlier rollout cannot change the chunk generated now. Concurrent test-time methods give a frozen policy such an input from retrieved successes, a learned critic, a verifier, or a dynamics model, but none uses the robot's own failed rollouts as negative evidence with nothing but a terminal outcome bit. We introduce TraceFlow, a progress-aligned guidance field that turns the action densities of retrieved successful and failed rollouts into a bounded correction to a frozen flow-matching action expert, using one terminal outcome bit per rollout and no other label. Its TraceBank stores traces, time-ordered state-action records with a terminal label, starts from the target-task training traces, and later admits the deployed robot's own rollouts. On an ordered real-robot packing task the base completes 21 of 50 trials in order, TraceFlow 39, and one stacking round without any weight update 47, with wrong-sequence episodes falling from 20 to 0. In simulation the gain is selective: with per-suite selected settings, TraceFlow raises RoboMemArena Sequence from 78.92\% to 91.50\% task success and Transferring from 54.41\% to 62.00\% at stacking round 2, leaves the 26-task aggregate unchanged, lowers Counting and Occlusion by 1.12 and 1.42 points, and changes LIBERO-Plus (Long) by +1.27 points (p = 0.0733). Stacking gains are finite, every branch peaking before round ten, and the bank's success-to-failure ratio predicts no retrieval allocation.
Abstract:A vision-language-action (VLA) policy with a flow-matching action expert generates each action chunk (a short command sequence) by integrating a learned velocity field; once its weights are fixed, the success or failure of an earlier rollout cannot change the chunk generated now. Concurrent test-time methods give a frozen policy such an input from retrieved successes, a learned critic, a verifier, or a dynamics model, but none uses the robot's own failed rollouts as negative evidence with nothing but a terminal outcome bit. We introduce TraceFlow, a progress-aligned guidance field that turns the action densities of retrieved successful and failed rollouts into a bounded correction to a frozen flow-matching action expert, using one terminal outcome bit per rollout and no other label. Its TraceBank stores traces, time-ordered state-action records with a terminal label, starts from the target-task training traces, and later admits the deployed robot's own rollouts. On an ordered real-robot packing task the base completes 21 of 50 trials in order, TraceFlow 39, and one stacking round without any weight update 47, with wrong-sequence episodes falling from 20 to 0. In simulation the gain is selective: with per-suite selected settings, TraceFlow raises RoboMemArena Sequence from 78.92\% to 91.50\% task success and Transferring from 54.41\% to 62.00\% at stacking round 2, leaves the 26-task aggregate unchanged, lowers Counting and Occlusion by 1.12 and 1.42 points, and changes LIBERO-Plus (Long) by +1.27 points (p = 0.0733). Stacking gains are finite, every branch peaking before round ten, and the bank's success-to-failure ratio predicts no retrieval allocation.
Robotics (cs.RO)
arXiv:2609.20646 [cs.RO]
(or arXiv:2609.20646v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.20646 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 · 346 words
Related coverage
Guided Action Flow: Q-Guided Inference for Flow-Matching Vision-Language-Action Policies
arXiv · Aug 25, 2026
Steering Generative Robot Policies with Lexicographic Preferences
arXiv · Sep 16, 2026
PreferenceFlow: Test-Time Guidance of Flow-Matching Robot Policies from Human Interventions
arXiv · Sep 30, 2026
Q-VGM: Q-Value-Gradient Matching for Off-Policy Reinforcement Learning of Flow-Matching VLA
arXiv · Jul 21, 2026