mlsanityData to 7 Oct 2026

See Less, Specify More: Visual Evidence Budgets for Generalizable VLAs

arXiv 2606.02735 ↗First posted 1 Jun 2026cs.RO cs.AI cs.LGcited by 1

On the map: Vision-language-action models → VLA architectures and generalization

Abstract (arXiv)

Generalization remains a central bottleneck for vision-language-action (VLA) models: under distractors, appearance shifts, and semantically similar tasks, the policy must often infer local execution details from coarse instructions while also deciding which parts of the image matter for control. We present S2 (See Less, Specify More), a framework for improving VLA generalization by training the executor under a cleaner interface. Specify More preserves the original instruction as a stable high-level goal while relabeling each trajectory into refined trajectory- and subtask-level language that disambiguates the current execution mode. Unlike native attention, See Less imposes an explicit visual evidence budget, training the executor to act from task-sufficient evidence rather than unconstrained visual context, without any region or mask annotation. This interface lets the executor follow detailed guidance without relying on distracting visual patches or resolving avoidable ambiguity on its own, and it remains compatible with off-the-shelf VLM planners through in-context learning. Across our main evaluation settings, S2 improves overall generalization metrics by changing the executor's learning problem: coarse instructions induce avoidable supervision aliasing, goal-preserving local guidance outperforms instruction replacement in our main ablations, and explicit evidence budgeting reduces dependence on broad visual context beyond efficiency considerations. Across eight real-robot tasks on TX-G2 (an AgiBot G2-compatible variant) and HSR, S2 raises mean subtask success from 54.2% to 79.0% over pi0.5. Together, these results suggest that VLA generalization improves when the executor is trained to act from informative local guidance and task-sufficient visual evidence, rather than recovering both from weak supervision.

Key facts. Cited by 1 paper in the map (data through 7 Oct 2026). Placed in VLA architectures and generalization, one of 13 directions in Vision-language-action models. Builds on 37 of its 44 references that are in the map. First posted on arXiv 1 Jun 2026.

Builds on

37 of 44 references.

References · 37

  1. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
  2. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  3. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  4. 2022Token Merging: Your ViT But Fastercited by 995
  5. 2021DynamicViT: Efficient Vision Transformers with Dynamic Token Sparsificationcited by 849
  6. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
  7. 2021CLIPort: What and Where Pathways for Robotic Manipulationcited by 617
  8. 2021CALVIN: A Benchmark for Language-Conditioned Policy Learning for Long-Horizon Robot Manipulation Taskscited by 564
  9. 2022Perceiver-Actor: A Multi-Task Transformer for Robotic Manipulationcited by 545
  10. 2022BC-Z: Zero-Shot Task Generalization with Robotic Imitation Learningcited by 517
  11. 2025SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Roboticscited by 507
  12. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 440
25 more
  1. 2022Not All Patches are What You Need: Expediting Vision Transformers via Token Reorganizationscited by 359
  2. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 285
  3. 2025Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Modelscited by 216
  4. 2025ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planningcited by 152
  5. 2025VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Modelcited by 140
  6. 2025Unified Vision-Language-Action Modelcited by 128
  7. 2021TokenLearner: What Can 8 Learned Tokens Do for Images and Videos?cited by 127
  8. 2025HAMSTER: Hierarchical Action Models For Open-World Robot Manipulationcited by 125
  9. 2022Robotic Skill Acquisition via Instruction Augmentation with Vision-Language Modelscited by 58
  10. 2017Deep Object-Centric Representations for Generalizable Robot Learningcited by 50
  11. 2025TGRPO :Fine-tuning Vision-Language-Action Model via Trajectory-wise Group Relative Policy Optimizationcited by 47
  12. 2025BitVLA: 1-bit Vision-Language-Action Models for Robotics Manipulationcited by 42
  13. 2026Steerable Vision-Language-Action Policies for Embodied Reasoning and Hierarchical Controlcited by 19
  14. 2024Policy Adaptation via Language Optimization: Decomposing Tasks for Few-Shot Imitationcited by 18
  15. 2024Object-Centric Instruction Augmentation for Robotic Manipulationcited by 16
  16. 2026Scaling World Model for Hierarchical Manipulation Policiescited by 15
  17. 2024Scaling Robot Policy Learning via Zero-Shot Labeling with Foundation Modelscited by 14
  18. 2024Leveraging Locality to Boost Sample Efficiency in Robotic Manipulationcited by 13
  19. 2024STEER: Flexible Robotic Manipulation via Dense Language Groundingcited by 13
  20. 2018Pay attention! - Robustifying a Deep Visuomotor Policy through Task-Focused Attentioncited by 11
  21. 2026ReSteer: Quantifying and Refining the Steerability of Multitask Robot Policiescited by 6
  22. 2026BFA++: Hierarchical Best-Feature-Aware Token Prune for Multi-View Vision Language Action Modelcited by 3
  23. 2026Spotlighting Task-Relevant Features: Object-Centric Representations for Better Generalization in Robotic Manipulationcited by 2
  24. 2024Efficient Training of Generalizable Visuomotor Policies via Control-Aware Augmentationcited by 1
  25. 2026Training-Free Interaction-Aligned Visual Token Pruning for Efficient Embodied Manipulationcited by 1

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026BICPO-VLA: Behavior-Identified Continuation Preference Optimization for Smooth Asynchronous Vision-Language-Action Controlcited by 1

Similar, not cited

Close in wording, no citation either way

  1. 2025Seeing to Act, Prompting to Specify: A Bayesian Factorization of Vision Language Action Policy
  2. 2025Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization
  3. 2025MAPS: Preserving Vision-Language Representations via Module-Wise Proximity Scheduling for Better Vision-Language-Action Generalization
  4. 2026Correcting WHERE, Preserving HOW: Compositional Generalization for Vision-Language-Action Models via Referential Guidance
  5. 2026VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies
  6. 2026DIVE: Dynamic Iterative Visual Evidence Construction for Efficient Vision-Language Models
  7. 2026FocusVLA: Focused Visual Utilization for Vision-Language-Action Models
  8. 2026VLA-InfoEntropy: A Training-Free Vision-Attention Information Entropy Approach for Vision-Language-Action Models Inference Acceleration and Success