mlsanityData to 7 Oct 2026

ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planning

arXiv 2507.16815 ↗First posted 22 Jul 2025cs.CV cs.AI cs.LGcited by 152

On the map: LLM reasoning → Spatial and embodied reasoning

also in 1 more
Abstract (arXiv)

Vision-language-action (VLA) reasoning tasks require agents to interpret multimodal instructions, perform long-horizon planning, and act adaptively in dynamic environments. Existing approaches typically train VLA models in an end-to-end fashion, directly mapping inputs to actions without explicit reasoning, which hinders their ability to plan over multiple steps or adapt to complex task variations. In this paper, we propose ThinkAct, a dual-system framework that bridges high-level reasoning with low-level action execution via reinforced visual latent planning. ThinkAct trains a multimodal LLM to generate embodied reasoning plans guided by reinforcing action-aligned visual rewards based on goal completion and trajectory consistency. These reasoning plans are compressed into a visual plan latent that conditions a downstream action model for robust action execution on target environments. Extensive experiments on embodied reasoning and robot manipulation benchmarks demonstrate that ThinkAct enables few-shot adaptation, long-horizon planning, and self-correction behaviors in complex embodied AI tasks.

Key facts. Cited by 152 papers in the map (data through 7 Oct 2026). Placed in Spatial and embodied reasoning, one of 17 directions in LLM reasoning. Builds on 50 of its 59 references that are in the map. First posted on arXiv 22 Jul 2025.

Builds on

50 of 59 references.

Closest ancestors · 6

Same direction, newest first

  1. 2025CoT-VLA: Visual Chain-of-Thought Reasoning for Vision-Language-Action Models27 Mar
  2. 2024EgoPlan-Bench2: A Benchmark for Multimodal Large Language Model Planning in Real-World Scenarios5 Dec
  3. 2024AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation1 Oct
  4. 2024Robotic Control via Embodied Chain-of-Thought Reasoning11 Jul
  5. 2023EgoPlan-Bench: Benchmarking Multimodal Large Language Models for Human-Level Planning11 Dec
  6. 2023RoboVQA: Multimodal Long-Horizon Reasoning for Robotics1 Nov

Other references · 44

  1. 2021Learning Transferable Visual Models From Natural Language Supervisioncited by 33,456
  2. 2023GPT-4 Technical Reportcited by 18,132
  3. 2022Chain-of-Thought Prompting Elicits Reasoning in Large Language Modelscited by 12,302
  4. 2024DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Modelscited by 8,851
  5. 2023Visual Instruction Tuningcited by 7,792
  6. 2023DINOv2: Learning Robust Visual Features without Supervisioncited by 7,764
  7. 2023BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Modelscited by 6,069
  8. 2025Qwen2.5-VL Technical Reportcited by 5,418
  9. 2024Gemini 1.5: Unlocking multimodal understanding across millions of tokens of contextcited by 3,175
  10. 2023RT-2: Vision-Language-Action Models Transfer Web Knowledge to Robotic Controlcited by 2,993
  11. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  12. 2023Diffusion Policy: Visuomotor Policy Learning via Action Diffusioncited by 2,730
32 more
  1. 2024LLaVA-OneVision: Easy Visual Task Transfercited by 2,669
  2. 2022RT-1: Robotics Transformer for Real-World Control at Scalecited by 2,007
  3. 2025InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Modelscited by 1,769
  4. 2024Expanding Performance Boundaries of Open-Source Multimodal Models with Model, Data, and Test-Time Scalingcited by 1,573
  5. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  6. 2023LIBERO: Benchmarking Knowledge Transfer for Lifelong Robot Learningcited by 1,338
  7. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  8. 2017The "something something" video database for learning and evaluating visual common sensecited by 1,092
  9. 2023Open X-Embodiment: Robotic Learning Datasets and RT-X Modelscited by 849
  10. 2024DROID: A Large-Scale In-The-Wild Robot Manipulation Datasetcited by 801
  11. 2023VILA: On Pre-training for Visual Language Modelscited by 686
  12. 2025Visual-RFT: Visual Reinforcement Fine-Tuningcited by 501
  13. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  14. 2025Video-R1: Reinforcing Video Reasoning in MLLMscited by 464
  15. 2025Demystifying Long Chain-of-Thought Reasoning in LLMscited by 323
  16. 2024TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policiescited by 298
  17. 2020Understanding Human Hands in Contact at Internet Scalecited by 281
  18. 2024RoboPoint: A Vision-Language Model for Spatial Affordance Prediction for Roboticscited by 242
  19. 2025Hi Robot: Open-Ended Instruction Following with Hierarchical Vision-Language-Action Modelscited by 216
  20. 2024Chain-of-Thought Reasoning Without Promptingcited by 208
  21. 2024NVILA: Efficient Frontier Visual Language Modelscited by 206
  22. 2023REFLECT: Summarizing Robot Experiences for Failure Explanation and Correctioncited by 171
  23. 2025Magma: A Foundation Model for Multimodal AI Agentscited by 138
  24. 2024Eagle: Exploring The Design Space for Multimodal LLMs with Mixture of Encoderscited by 137
  25. 2025HAMSTER: Hierarchical Action Models For Open-World Robot Manipulationcited by 125
  26. 2024LLARVA: Vision-Action Instruction Tuning Enhances Robot Learningcited by 87
  27. 2025Eagle 2: Building Post-Training Data Strategies from Scratch for Frontier Vision-Language Modelscited by 69
  28. 2025Eagle 2.5: Boosting Long-Context Post-Training for Frontier Vision-Language Modelscited by 62
  29. 2024LLaRA: Supercharging Robot Learning Data for Vision-Language Policycited by 51
  30. 2024From Multimodal LLMs to Generalist Embodied Agents: Methods and Lessonscited by 42
  31. 2024In-Context Learning Enables Robot Action Prediction in LLMscited by 25
  32. 2025Action-Free Reasoning for Policy Generalizationcited by 24

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2025: 39; 2026: 113.

152 papers.

Sort:

By area:

  1. 2026Causal World Modeling for Robot Controlcited by 289
  2. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 285
  3. 2025MolmoAct: Action Reasoning Models that can Reason in Spacecited by 189
  4. 2025LIBERO-PRO: Towards Robust and Fair Evaluation of Vision-Language-Action Models Beyond Memorizationcited by 166
  5. 2025VLA-Adapter: An Effective Paradigm for Tiny-Scale Vision-Language-Action Modelcited by 140
  6. 2025mimic-video: Video-Action Models for Generalizable Robot Control Beyond VLAscited by 133
  7. 2025RynnVLA-002: A Unified Vision-Language-Action and World Modelcited by 70
  8. 2025FlowVLA: Visual Chain of Thought-based Motion Reasoning for Vision-Language-Action Modelscited by 50
  9. 2025Pure Vision Language Action (VLA) Models: A Comprehensive Surveycited by 46
  10. 2026MolmoAct2: Action Reasoning Models for Real-world Deploymentcited by 45
  11. 2026Xiaomi-Robotics-0: An Open-Sourced Vision-Language-Action Model with Real-Time Executioncited by 42
  12. 2026VLM4VLA: Revisiting Vision-Language-Models in Vision-Language-Action Modelscited by 38
38 more
  1. 2025Unified Diffusion VLA: Vision-Language-Action Model via Joint Discrete Denoising Diffusion Processcited by 37
  2. 2025Embodied-R1: Reinforced Embodied Reasoning for General Robotic Manipulationcited by 36
  3. 2025Planning with Reasoning using Vision Language World Modelcited by 36
  4. 2025FailSafe: Reasoning and Recovery from Failures in Vision-Language-Action Modelscited by 33
  5. 2025EO-1: An Open Unified Embodied Foundation Model for General Robot Controlcited by 30
  6. 2025HiF-VLA: Hindsight, Insight and Foresight through Motion Representation for Vision-Language-Action Modelscited by 30
  7. 2026ACoT-VLA: Action Chain-of-Thought for Vision-Language-Action Modelscited by 30
  8. 2026LaST0: Latent Spatio-Temporal Chain-of-Thought for Robotic Vision-Language-Action Modelcited by 28
  9. 2025NORA-1.5: A Vision-Language-Action Model Trained using World Model- and Action-based Preference Rewardscited by 26
  10. 2025Do What You Say: Steering Vision-Language-Action Models via Runtime Reasoning-Action Alignment Verificationcited by 25
  11. 2025SRPO: Self-Referential Policy Optimization for Vision-Language-Action Modelscited by 25
  12. 2025Reasoning Within the Mind: Dynamic Multimodal Interleaving in Latent Spacecited by 25
  13. 2026OA-WAM: Object-Addressable World Action Model for Robust Robot Manipulationcited by 25
  14. 2025Robobench: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models as Embodied Braincited by 24
  15. 2025FPC-VLA: A Vision-Language-Action Framework with a Supervisor for Failure Prediction and Correctioncited by 21
  16. 2025Counterfactual VLA: Self-Reflective Vision-Language-Action Model with Adaptive Reasoningcited by 21
  17. 2026AtomVLA: Scalable Post-Training for Robotic Manipulation via Predictive Latent World Modelscited by 21
  18. 2026Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planningcited by 20
  19. 2026DualCoT-VLA: Visual-Linguistic Chain of Thought via Parallel Reasoning for Vision-Language-Action Modelscited by 20
  20. 2026Latent Reasoning VLA: Latent Thinking and Prediction for Vision-Language-Action Modelscited by 19
  21. 2026Recurrent-Depth VLA: Implicit Test-Time Compute Scaling of Vision-Language-Action Models via Latent Iterative Reasoningcited by 17
  22. 2025AsyncVLA: Asynchronous Flow Matching for Vision-Language-Action Modelscited by 16
  23. 2025Bridging the Semantic-Action Gap in Visual Token Pruning for Efficient VLA Inferencecited by 14
  24. 2026TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformerscited by 14
  25. 2026CycleVLA: Proactive Self-Correcting Vision-Language-Action Models via Subtask Backtracking and Minimum Bayes Risk Decodingcited by 13
  26. 2025Towards Spatial Trace with Reasoning in Vision-Language Models for Roboticscited by 12
  27. 2026VLANeXt: Recipes for Building Strong VLA Modelscited by 12
  28. 2025PEEK: Guiding and Minimal Image Representations for Zero-Shot Generalization of Robot Manipulation Policiescited by 11
  29. 2025DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Actioncited by 11
  30. 2026Towards Long-Lived Robots: Continual Learning VLA Models via Reinforcement Fine-Tuningcited by 11
  31. 2026ROCKET: Residual-Oriented Multi-Layer Alignment for Spatially-Aware Vision-Language-Action Modelscited by 11
  32. 2026LaMP: Learning Vision-Language-Action Policy with 3D Scene Flow as Latent Motion Priorcited by 11
  33. 2025Spatial-Aware VLA Pretraining through Visual-Physical Alignment from Human Videoscited by 10
  34. 2026VLingNav: Embodied Navigation with Adaptive Reasoning and Visual-Assisted Linguistic Memorycited by 10
  35. 2026MetricAnything: Scaling Metric Depth Pretraining with Noisy Heterogeneous Sourcescited by 10
  36. 2026SimVLA: A Simple VLA Baseline for Robotic Manipulationcited by 10
  37. 2026Fast-dVLA: Accelerating Discrete Diffusion VLA to Real-Time Performancecited by 9
  38. 2026Goal2Skill: Long-Horizon Manipulation with Adaptive Planning and Reflectioncited by 9

50 most cited of 152 All 152 (JSON)

Similar, not cited

Close in wording, no citation either way

  1. 2026PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
  2. 2026Afford-VLA: Action-Aligned Visual Planning via Internalized Affordance
  3. 2026StreamVLA: Breaking the Reason-Act Cycle via Completion-State Gating
  4. 2026VisualThink-VLA: Visual Intermediate Reasoning for Effective and Low-Latency Vision-Language-Action Policies
  5. 2026PHR-VLA: Planning Horizon Reasoning for Vision-Language-Action Models
  6. 2026RoboAgent: Chaining Basic Capabilities for Embodied Task Planning
  7. 2026ThinkingVLA: Interleaved Vision and Language Reasoning for Robotic Manipulation
  8. 2025LoHoVLA: A Unified Vision-Language-Action Model for Long-Horizon Embodied Tasks