mlsanityData to 14 Oct 2026

Test-Time Perturbation Learning with Delayed Feedback for Vision-Language-Action Models

arXiv 2604.18107 ↗First posted 20 Apr 2026cs.CVcited by 0

On the map: Vision-language-action models → Efficient VLA inference and adaptation

Abstract (arXiv)

Vision-Language-Action models (VLAs) achieve remarkable performance in sequential decision-making but remain fragile to subtle environmental shifts, such as small changes in object pose. We attribute this brittleness to trajectory overfitting, where VLAs over-attend to the spurious correlation between actions and entities, then reproduce memorized action patterns. We propose Perturbation learning with Delayed Feedback (PDF), a verifier-free test-time adaptation framework that improves decision performance without fine-tuning the base model. PDF mitigates the spurious correlation through uncertainty-based data augmentation and action voting, while an adaptive scheduler allocates augmentation budgets to balance performance and efficiency. To further improve stability, PDF learns a lightweight perturbation module that retrospectively adjusts action logits guided by delayed feedback, correcting overconfidence issue. Experiments on LIBERO (+7.4% success rate) and Atari (+10.3 human normalized score) demonstrate consistent gains of PDF in task success over vanilla VLA and VLA with test-time adaptation, establishing a practical path toward reliable test-time adaptation in multimodal decision-making agents. The code is available at hrefhttps://github.com/zhoujiahuan1991/CVPR2026-PDFhttps://github.com/zhoujiahuan1991/CVPR2026-PDF.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Efficient VLA inference and adaptation, one of 13 directions in Vision-language-action models. Builds on 27 of its 33 references that are in the map. First posted on arXiv 20 Apr 2026.

Builds on

27 of 33 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Verifier-free Test-Time Sampling for Vision-Language-Action Models7 Oct
  2. 2025RoboMonkey: Scaling Test-Time Sampling and Verification for Vision-Language-Action Models21 Jun

Other references · 25

  1. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 3,080
  2. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,838
  3. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,407
  4. 2024Octo: An Open-Source Generalist Robot Policycited by 1,484
  5. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,410
  6. 2017PackNet: Adding Multiple Tasks to a Single Network by Iterative Pruningcited by 898
  7. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 883
  8. 2022A Generalist Agentcited by 808
  9. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 519
  10. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 451
  11. 2023An Embodied Generalist Agent in 3D Worldcited by 365
  12. 2023Vision-Language Foundation Models as Effective Robot Imitatorscited by 349
13 more
  1. 20243D-VLA: A 3D Vision-Language-Action Generative World Modelcited by 328
  2. 2024TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policiescited by 304
  3. 2025LIBERO-Plus: In-depth Robustness Analysis of Vision-Language-Action Modelscited by 303
  4. 2023Any-point Trajectory Modeling for Policy Learningcited by 215
  5. 2025SimpleVLA-RL: Scaling VLA Training via Reinforcement Learningcited by 151
  6. 2024Steering Your Generalists: Improving Robotic Foundation Models via Value Guidancecited by 106
  7. 2025TGRPO :Fine-tuning Vision-Language-Action Model via Trajectory-wise Group Relative Policy Optimizationcited by 49
  8. 2025JARVIS-VLA: Post-Training Large-Scale Vision Language Models to Play Visual Games with Keyboards and Mousecited by 28
  9. 2024Jack of All Trades, Master of Some, a Multi-Purpose Transformer Agentcited by 21
  10. 2025SCAP: Transductive Test-Time Adaptation via Supportive Clique-based Attribute Promptingcited by 12
  11. 2025MetaVLA: Unified Meta Co-training For Efficient Embodied Adaptioncited by 5
  12. 2025Test-Time Adaptation with Binary Feedbackcited by 3
  13. 2025Class-aware Domain Knowledge Fusion and Fission for Continual Test-Time Adaptationcited by 3

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2025Steering Vision-Language-Action Models as Anti-Exploration: A Test-Time Scaling Approach
  2. 2026VANE: Reliable Test-Time Training for Vision-Language-Action Models via Future Visual Representation Prediction
  3. 2026PerturBot: Breaking Shortcut Priors in Vision-Language-Action Models with Perturbative Training
  4. 2026Verify, Act, Distill: Turning Test-Time Compute into Test-Time Adaptation for Vision-Language-Action Models
  5. 2026VLA-ATTC: Adaptive Test-Time Compute for VLA Models with Relative Action Critic Model
  6. 2025EVOLVE-VLA: Test-Time Training from Environment Feedback for Vision-Language-Action Models
  7. 2026PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models
  8. 2026A Causality-aware Infer-diagnose-refine Framework for Test-time Modality Adaptation in VLA Models