Retrieve in Time, Correct in Frequency
Retrieve in Time, Correct in Frequency
The paper presents RTCF, a training-free test-time correction framework for frozen vision-language-action (VLA) policies. It improves performance by aligning visual execution history with successful trajectories and transferring low-frequency residuals. RTCF achieves higher success rates in long-horizon manipulation tasks without requiring parameter updates or additional GPU resources.
This paper introduces RTCF, a test-time correction framework for frozen vision-language-action policies, which improves performance without requiring parameter updates or additional computational resources. The method uses progressive memory alignment and transfers low-frequency residuals to enhance success rates in long-horizon manipulation tasks.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
median latencies
10.99
View original evidence
2,000 episodes per condition, RTCF raises aggregate success from 86.4% to 88.4% and improves LIBERO-Long from 61.6% to 68.6%.These gains require no parameter updates, repeated VLA inference, or additional GPU resources: correction can be performed on the client CPU after a single policy invocation, and the median latencies sum to only 10.99 ms per action chunk
Open source S5episodes per condition
2,000 episodes
View original evidence
2,000 episodes per condition, RTCF raises aggregate success from 86.4% to 88.4% and improves LIBERO-Long from 61.6% to 68.6%.These gains require no parameter updates, repeated VLA inference, or additional GPU resources: correction can be performed on the client CPU after a single policy invocation, and the median latencies sum to only 10.99 ms per action chunk
Open source S5
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
2,000 episodes per condition, RTCF raises aggregate success from 86.4% to 88.4% and improves LIBERO-Long from 61.6% to 68.6%.These gains require no parameter updates, repeated VLA inference, or additional GPU resources: correction can be performed on the client CPU after a single policy invocation, and the median latencies sum to only 10.99 ms per action chunk
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org