mlsanityData to 7 Oct 2026

SCULPT-VLA: Learning Structured Control through Staged Action Grounding

arXiv 2609.23275 ↗First posted 20 Sep 2026cs.ROcited by 0

On the map: Vision-language-action models → Human-video and latent-action pretraining also fits: Reasoning and chain-of-thought in VLA

Abstract (arXiv)

Vision-language-action (VLA) policies increasingly incorporate structured intermediate supervision beyond action labels. Yet specifying what an intermediate representation should encode leaves open how action prediction learns to depend on it. We introduce SCULPT-VLA, a policy that learns structured control through staged action grounding. Its action-conditioning state comprises complementary factors for task progression, scene dynamics, and spatial grounding. Training first forms these factors with teacher scaffolds, then grounds coarse action prediction through their composition as scaffold inputs are withdrawn. Direct perceptual access is subsequently restored for continuous refinement, combining the learned state with perceptual detail. The curriculum separates learning to condition actions on structure from refining continuous control. Deployment requires neither teachers nor discrete-action autoregression. SCULPT-VLA achieves higher average success than shared-backbone baselines on LIBERO, SimplerEnv-WidowX, and RoboTwin 2.0 Full. On SimplerEnv-WidowX, final success is 83.5%, versus 71.3% when Stage-II action learning directly accesses vision and language. Across four physical robot tasks, average success under the tested distribution shifts reaches 58.1%, compared with 45.6% for π0.5. Training ablations and factor-wise interventions support the staged design and show that the learned state continues to contribute to control after direct perceptual access is restored.

Key facts. No citing papers in the map yet (data through 7 Oct 2026). Placed in Human-video and latent-action pretraining, one of 13 directions in Vision-language-action models. Builds on 54 of its 54 references that are in the map. First posted on arXiv 20 Sep 2026.

Builds on

54 of 54 references.

Closest ancestors · 2

Same direction, newest first

  1. 2025Seeing Space and Motion: Enhancing Latent Actions with Geometric and Dynamic Awareness for Vision-Language-Action Models30 Sep
  2. 2025UniVLA: Learning to Act Anywhere with Task-centric Latent Actions9 May

Other references · 52

  1. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  2. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  3. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  4. 2025Qwen3-VL Technical Reportcited by 2,534
  5. 2024π0: A Vision-Language-Action Flow Model for General Robot Controlcited by 2,297
  6. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  7. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  8. 2024Depth Anything V2cited by 1,385
  9. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  10. 2025Qwen3 Embedding: Advancing Text Embedding and Reranking Through Foundation Modelscited by 1,175
  11. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  12. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Successcited by 844
38 more
  1. 2023BridgeData V2: A Dataset for Robot Learning at Scalecited by 634
  2. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 502
  3. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  4. 2025RoboTwin 2.0: A Scalable Data Generator and Benchmark with Strong Domain Randomization for Robust Bimanual Robotic Manipulationcited by 477
  5. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Modelscited by 440
  6. 2024Robotic Control via Embodied Chain-of-Thought Reasoningcited by 340
  7. 2026Fast-WAM: Do World Action Models Need Test-time Future Imagination?cited by 311
  8. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 285
  9. 2025Motus: A Unified Latent Action World Modelcited by 231
  10. 2025DreamVLA: A Vision-Language-Action Model Dreamed with Comprehensive World Knowledgecited by 203
  11. 2025ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planningcited by 152
  12. 2025VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Modelcited by 140
  13. 2025Unified Vision-Language-Action Modelcited by 128
  14. 2026VLA-JEPA: Enhancing Vision-Language-Action Model with Latent World Modelcited by 104
  15. 2026StarVLA: A Lego-like Codebase for Vision-Language-Action Model Developingcited by 103
  16. 2026A Pragmatic VLA Foundation Modelcited by 97
  17. 2025F1: A Vision-Language-Action Model Bridging Understanding and Generation to Actionscited by 85
  18. 2025BridgeVLA: Input-Output Alignment for Efficient 3D Manipulation Learning with Vision-Language Modelscited by 77
  19. 2025InstructVLA: Vision-Language-Action Instruction Tuning from Understanding to Manipulationcited by 71
  20. 2025FlowVLA: Visual Chain of Thought-based Motion Reasoning for Vision-Language-Action Modelscited by 50
  21. 2024RT-Affordance: Affordances are Versatile Intermediate Representations for Robot Manipulationcited by 40
  22. 2025Unified Diffusion VLA: Vision-Language-Action Model via Joint Discrete Denoising Diffusion Processcited by 37
  23. 2026LaST0: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Modelcited by 28
  24. 2026When Vision Overrides Language: Evaluating and Mitigating Counterfactual Failures in VLAscited by 27
  25. 2026LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queriescited by 24
  26. 2026Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planningcited by 20
  27. 2026Latent Reasoning VLA: Latent Thinking and Prediction for Vision-Language-Action Modelscited by 19
  28. 2026OpenWAM: An Open, Modular Exploration Towards Systematic World-Action Model Pretrainingcited by 17
  29. 2026Not All Features Are Created Equal: A Mechanistic Study of Vision-Language-Action Modelscited by 14
  30. 2026LaMP: Learning Vision-Language-Action Policy with 3D Scene Flow as Latent Motion Priorcited by 11
  31. 2026DeltaVLA: Prior-Guided Vision-Language-Action Models via World Knowledge Variationcited by 9
  32. 2025LoLA: Long Horizon Latent Action Learning for General Robot Manipulationcited by 8
  33. 2026Chain of World: World Model Thinking in Latent Motioncited by 8
  34. 2026Making Foresight Actionable: Repurposing Representation Alignment in World Action Modelscited by 8
  35. 2026VLA-Thinker: Boosting Vision-Language-Action Models through Thinking-with-Image Reasoningcited by 5
  36. 2026Drop-Then-Recovery: How Redundant Are Vision-Language-Action Models?cited by 3
  37. 2026Bridge-WA: Predicting Where and How the World Changes for Robotic Actioncited by 2
  38. 2026GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulationcited by 2

50 most cited of 52 All 52 (JSON)

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026ActionGround: Training-Free Runtime Refinement of Frozen VLA Policies
  2. 2026LayerRoute: Action-Conditioned Mixture-of-Layers Routing for Vision-Language-Action Policies
  3. 2026CAC-VLA: Context-Gated Action Conditioning for Vision-Language-Action Models
  4. 2026AC-VLA: Robust Out-of-Distribution Action Execution via Compositional Learning
  5. 2026Two Bridges, One Pathway: From VLMs to Generalizable VLAs with Embodied Trajectory-Coupled Data
  6. 2026Mitigating State Aliasing in Vision-Language-Action Models via Inverse Dynamics Learning
  7. 2026Reflective VLA: In-Context Action Consequences Make VLAs Generalize
  8. 2026LA4VLA: Learning to Act without Seeing via Language-Action Pretraining