cs.RONov 16, 2026

ReRead: Wrong Predictions, Useful Representations - Correcting Frozen Robot Policies by Re-reading Their Own Representations

Authors: Xu Su, Shunlong Xiao, Qintao Yi, Mingkuan Zhao, Xinmiao Yu, Yizhe Li, Zihan Wang, Bokui Chen

Organizations: Tsinghua University · Xi’an Jiaotong University

Abstract

When a pretrained robot policy fails on challenging tasks, common remedies include retraining the backbone, adding an external visual encoder, or introducing a memory module. We argue that wrong predictions do not imply wrong representations: a policy can retain the required information yet fail to read it out as the correct action. To test this view, we propose ReRead: with the backbone frozen, a corrector re-reads the current and two previous caches to predict an action residual. The corrector sees no images, is trained solely on offline demonstrations from all 50 RoboTwin 2.0 tasks, and adds only 4.17% latency. Ablations and interventions show that the correction relies on the backbone's own representations: none of four external visual encoders adds further gain, and swapping in past caches from other trajectories sharply increases action error. On the 11 hardest RoboTwin 2.0 tasks, ReRead improves the success rates of LingBot-VLA, π0.5π_{0.5}, and Fast-WAM by 23.45, 13.09, and 8.00 percentage points, respectively. In paired evaluation on all 50 tasks, it raises Fast-WAM's average success rate from 90.73% to 92.66%. On a Flexiv dual-arm robot, it raises success on a phase-ambiguous task from 0% to 36%.

Figures & tables

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

Nov 9, 2026

Distilling Instructions into Weights: Closed-Form Edits Fix a Robot Policy's Misreading

Generalist robot policies often fail at one step of a task they can otherwise perform. We ask whether such a failure can be repaired by editing the policy's weights directly, without gradients, training data, or a change to the instruction. Told to put the cube then the banana in the bowl, a zero-shot π0.5π_{0.5} vision-language-action policy in RoboLab releases the cube before it reaches the bowl. Replaying 21 of these decisions under rewordings shows that it binds "in the bowl" to the wrong object. Closed-form instruction distillation copies the per-layer effect of an explicit rewording into the language model's MLP weights. On 128 held-out seeds, success rises from 27 to 106 episodes, with 81 fixed and 2 broken (p=7×10−22p=7\times10^{-22}), while an update of the same size in random directions leaves success unchanged. The edit also lifts two other tasks without detectable harm on the remaining 13, and a variant that stores 0.16% of the parameter count keeps most of its effect. Placed in the action expert or action head instead, the same distillation helps less or not at all. Our released tool reproduced the fix end to end on new held-out seeds, from 8/64 to 36/64. Edits aimed at perception, a missing grasp, contact physics, or the gripper threshold did not help, and of seven blind Claude Opus editors, only the one given the misread instruction produced an edit that passed held-out scoring. These results suggest that weight edits work best when a known-good input already exists to distil. Because reruns with unchanged weights flip about a third of paired outcomes, evaluations should pair seeds, rerun the unedited policy, and confirm gains on held-out seeds.
Oct 2, 2026

REDIRECT: A 1% Fix for Bad Robot Habits

Robots can acquire bad habits from a few defective moments in otherwise useful teleoperation. In mixed-quality robot data, normal and problematic demonstrations share most task behavior and differ only at a local action continuation. Full retraining is costly, while fine-tuning on clean data alone offers limited recovery under a small update budget. We ask whether the local difference can instead be fixed with only 1% of the full-training sample-backward budget. Using only episode-level retained/problematic labels, REDIRECT localizes the branch, assigns a coherent retained continuation to the original observations around it, and anchors shared behavior, without frame-level annotations or additional interaction. Across three randomized ManiSkill tasks and three seeds, REDIRECT raises the mean success rate from 68.2% to 90.3%, recovering 88.8% of the clean-retraining gap versus 22.1% for matched-compute fine-tuning. On a PiPER arm, it recovers 86.7-92.9% of the clean-only gap across cup insertion and towel folding. Local robot habits can therefore be repaired by spending the update budget on the behavioral difference rather than relearning shared behavior.
Oct 6, 2026

Compact Robot Policies Need Fine-Grained Visual Representations

Multi-task manipulation policies differ in architecture, scale, and pretrained priors all at once, so published comparisons cannot attribute performance to any single component. We argue that most of it comes from the visual representation, and that parameter scale and generative priors are largely incidental. To test this, we build CoRP (Compressed Representation Policy), a deliberately compact policy (48.9M parameters, no vision-language model and no video-generative prior) that factorizes into a representation extractor and a flow-matching action generator. It reaches 97.0% on LIBERO and 75.78%/73.36% on RoboTwin 2.0 Clean/Randomized, matching systems 40.9-163.6x larger. Holding the action generator fixed, we then vary one extractor property at a time. Pretrained initialization is decisive: a random ViT-S/14 drops to 78.1% and an ImageNet ResNet-34 to 74.5% on LIBERO. Pretraining alone is not enough, as freezing the encoder costs 19.8 points. Compression matters as much: resampling each view to 48 tokens beats passing all patch tokens (97.0% vs 83.2%), and a variational information bottleneck over those tokens is worse than a hard token budget, cutting LIBERO-Goal from 95.8% to 33.0% by suppressing the instruction-dependent token selection the policy relies on. Language conditioning contributes only where the observation leaves the goal ambiguous (LIBERO-Goal: 9.2% to 95.8%), while on RoboTwin 2.0, where observations are unambiguous, removing it slightly improves success. Therefore, we argue that a compact policy works when its representation is pretrained, task-adapted, and compressed. Project page: https://corp-policy.github.io/