VLA Policies Do Not Know When Not to Act: Unintended Task Execution and Start Suppression
cs.RO, cs.AI
Submitted: 2026-10-30
Updated: 2026-10-30
Comments: 17 pages, 3 figures, 13 tables. Code: https://github.com/zebehn/vla-start-suppression
Code: https://github.com/zebehn/vla-start-suppression
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Robust Skills, Brittle Grounding: Diagnosing Restricted Generalization in Vision-Language Action Policies via Multi-Object Picking
- When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAs
- Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Models
- ConflictVLA-Bench: Benchmarking Behavioral Responses of Vision-Language-Action Models to Premise Conflicts
- LangGap: Diagnosing and Closing the Language Gap in Vision-Language-Action Models
- Do What? Teaching Vision-Language-Action Models to Reject the Impossible
- Act, Think or Abstain: Complexity-Aware Adaptive Inference for Vision-Language-Action Models
- Encoded but Not in Control: Revealing the Grounding Gap in Vision-Language Robot Policies
- What the Guard Misses, the Robot Executes: Implied Harm in VLA Instructions
- Proactive Robot Assistance via Event-Driven Reasoning over State Changes
- GR00T N1: An Open Foundation Model for Generalist Humanoid Robots
- Beyond Single-Axis Testing: Paired Evaluation of Compound Robustness in Vision-Language-Action Policies
- SmolVLA: A Vision-Language-Action Model for Affordable and Efficient Robotics
- BOKBO (Best of K Bad Options): Calibrated Abstention for VLA Policies
- RationalVLA: A Rational Vision-Language-Action Model with Dual System
- Where Should Language Sit in a Multimodal Model? Lessons from What Language Does to Human Perception and Cognition
- Learning to Assist: Collaborative VLAs for Implicit Human-Robot Collaboration
- The Yes-Man Syndrome: Benchmarking Abstention in Embodied Robotic Agents
- Fast-WAM: Do World Action Models Need Test-time Future Imagination?
- Restoring Linguistic Grounding in VLA Models via Train-Free Attention Recalibration
Related papers
- Rethinking Actions: Continuity-Aware Priors for World-Model Planning
- MOSAIC: Modular Obstacle and Spatial-Aware Framework for Manipulation In Cluttered Scenes
- SAGAS: Semantic-Aware Graph-Assisted Stitching for Offline Temporal Logic Planning
- DexSeed: Expanding Human Hand-Object Interactions into Robot Demonstrations for Dexterous Policy Learning
- LaST0.5: A Generalist World Action Model with Interaction-Centric Latent Reasoning
- CIALLO: Learning Shared Intent Representations for Cross-Embodiment Manipulation