mlsanityData to 7 Oct 2026

TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers

arXiv 2601.14133 ↗First posted 20 Jan 2026cs.RO cs.CVcited by 14

On the map: Multimodal LLMs / vision-language models → Embodied and 3D vision-language models

also in 1 more
Abstract (arXiv)

The fundamental premise of Vision-Language-Action (VLA) models is to harness the extensive general capabilities of pre-trained Vision-Language Models (VLMs) for generalized embodied intelligence. However, standard robotic fine-tuning inevitably disrupts the pre-trained feature space, leading to "catastrophic forgetting" that compromises the general visual understanding we aim to leverage. To effectively utilize the uncorrupted general capabilities of VLMs for robotic tasks, we propose TwinBrainVLA, which coordinates two isomorphic VLM pathways: a frozen generalist (also called "Left Brain") and a trainable specialist (also called "Right Brain"). Our architecture utilizes a Asymmetric Mixture-of-Transformers (AsyMoT) mechanism, enabling the Right Brain to dynamically query and fuse intact semantic knowledge from the Left Brain with proprioceptive states. This fused representation conditions a flow-matching action expert for precise continuous control. Empirical results on SimplerEnv and RoboCasa benchmarks demonstrate that by explicitly retaining general capabilities, TwinBrainVLA achieves substantial performance gains over baseline models in complex manipulation tasks.

Key facts. Cited by 14 papers in the map (data through 7 Oct 2026). Placed in Embodied and 3D vision-language models, one of 16 directions in Multimodal LLMs / vision-language models. Builds on 43 of its 55 references that are in the map. First posted on arXiv 20 Jan 2026.

Builds on

43 of 55 references.

Closest ancestors · 12

Same direction, newest first

  1. 2026VLM4VLA: Revisiting Vision-Language-Models in Vision-Language-Action Models6 Jan
  2. 2025PhysBrain: Human Egocentric Data as a Bridge from Vision Language Models to Physical Intelligence18 Dec
  3. 2025XR-1: Towards Versatile Vision-Language-Action Models via Learning Unified Vision-Motion Representations4 Nov
  4. 2025Don't Blind Your VLA: Aligning Visual Representations for OOD Generalization29 Oct
  5. 2025Actions as Language: Fine-Tuning VLMs into VLAs Without Catastrophic Forgetting26 Sep
  6. 2025Igniting VLMs toward the Embodied Space15 Sep
  7. 2025RynnEC: Bringing MLLMs into Embodied World19 Aug
  8. 2025RoboRefer: Towards Spatial Referring with Reasoning in Vision-Language Models for Robotics4 Jun
  9. 2025Why Is Spatial Reasoning Hard for VLMs? An Attention Mechanism Perspective on Focus Areas3 Mar
  10. 2025Fine-Tuning Vision-Language-Action Models: Optimizing Speed and Success27 Feb
  11. 2024What Matters in Building Vision-Language-Action Models for Generalist Robots18 Dec
  12. 2024LLaVA-3D: A Simple yet Effective Pathway to Empowering LMMs with 3D-awareness26 Sep

Other references · 31

  1. 2024OpenVLA: An Open-Source Vision-Language-Action Modelcited by 2,777
  2. 2022Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answeringcited by 1,881
  3. 2025π0.5: a Vision-Language-Action Model with Open-World Generalizationcited by 1,653
  4. 2024Octo: An Open-Source Generalist Robot Policycited by 1,427
  5. 2025GR00T N1: An Open Foundation Model for Generalist Humanoid Robotscited by 1,167
  6. 2023SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehensioncited by 929
  7. 2025SpatialVLA: Exploring Spatial Representations for Visual-Language-Action Modelcited by 502
  8. 2024Evaluating Real-World Robot Manipulation Policies in Simulationcited by 484
  9. 2024CogACT: A Foundational Vision-Language-Action Model for Synergizing Cognition and Action in Robotic Manipulationcited by 411
  10. 2024RoboCasa: Large-Scale Simulation of Everyday Tasks for Generalist Robotscited by 391
  11. 2024TraceVLA: Visual Trace Prompting Enhances Spatial-Temporal Awareness for Generalist Robotic Policiescited by 298
  12. 2025X-VLA: Soft-Prompted Transformer as Scalable Cross-Embodiment Vision-Language-Action Modelcited by 285
19 more
  1. 2025MolmoAct: Action Reasoning Models that can Reason in Spacecited by 189
  2. 2025ThinkAct: Vision-Language-Action Reasoning via Reinforced Visual Latent Planningcited by 152
  3. 2025Spatial Forcing: Implicit Spatial Representation Alignment for Vision-language-action Modelcited by 125
  4. 2025Cambrian-S: Towards Spatial Supersensing in Videocited by 123
  5. 2025ChatVLA: Unified Multimodal Understanding and Robot Control with Vision-Language-Action Modelcited by 112
  6. 2025Galaxea Open-World Dataset and G0 Dual-System VLA Modelcited by 90
  7. 2025Universal Actions for Enhanced Embodied Foundation Modelscited by 80
  8. 2025InstructVLA: Vision-Language-Action Instruction Tuning from Understanding to Manipulationcited by 71
  9. 2025VideoVLA: Video Generators Can Be Generalizable Robot Manipulatorscited by 61
  10. 2025VLA-R1: Enhancing Reasoning in Vision-Language-Action Modelscited by 38
  11. 2025Evo-1: Lightweight Vision-Language-Action Model with Preserved Semantic Alignmentcited by 35
  12. 2025EO-1: An Open Unified Embodied Foundation Model for General Robot Controlcited by 30
  13. 2025DSI-Bench: A Benchmark for Dynamic Spatial Intelligencecited by 20
  14. 2026Fast-ThinkAct: Efficient Vision-Language-Action Reasoning via Verbalizable Latent Planningcited by 20
  15. 2025DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Actioncited by 11
  16. 2025SpatialTree: How Spatial Abilities Branch Out in MLLMscited by 11
  17. 2025How Do VLAs Effectively Inherit from VLMs?cited by 6
  18. 2026Limited Linguistic Diversity in Embodied AI Datasetscited by 5
  19. 2025Hybrid Training for Vision-Language-Action Modelscited by 2

Built on by

Cited by year (2026 to 7 Oct)

Citing papers per year: 2026: 14.

14 papers.

Sort:

By area:

  1. 2026LangForce: Bayesian Decomposition of Vision Language Action Models via Latent Action Queriescited by 24
  2. 2026Learning Native Continuation for Action Chunking Flow Policiescited by 18
  3. 2026Evo-Depth: A Lightweight Depth-Enhanced Vision-Language-Action Modelcited by 13
  4. 2026IntentVLA: Short-Horizon Intent Modeling for Aliased Robot Manipulationcited by 7
  5. 2026FrameSkip: Learning from Fewer but More Informative Frames in VLA Trainingcited by 6
  6. 2026UAM: A Dual-Stream Perspective on Forgetting in VLA Trainingcited by 5
  7. 2026PhysBrain 1.0 Technical Reportcited by 2
  8. 2026Latent Action as Intention Enables Efficient Future Imagination for World Action Modelscited by 2
  9. 2026GIFT: Guided Intermediate Feature Training via Action-Oriented Structural Supervision for Robotic Manipulationcited by 2
  10. 2026VLA-GSE: Boosting Parameter-Efficient Fine-Tuning in VLA with Generalized and Specialized Expertscited by 1
  11. 2026How Should Vision-Language-Action Models Use Proprioceptive State?cited by 1
  12. 2026In-Context VLA: Endowing Vision-Language-Action Models with Language via In-Context Post-Training and Agentic Tool Usecited by 1
2 more
  1. 2026RoboSemanticBench: Diagnosing Semantic Grounding in Action Prediction for VLA Modelscited by 0
  2. 2026Think Like a World Model, Act Like a VLA: Distilling World-Model Representations into Compact Robot Policiescited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026M2-VLA: Boosting Vision-Language Models for Generalizable Manipulation via Layer Mixture and Meta-Skills
  2. 2025MoLe-VLA: Dynamic Layer-skipping Vision Language Action Model via Mixture-of-Layers for Efficient Robot Manipulation
  3. 2026Libra-VLA: Achieving Learning Equilibrium via Asynchronous Coarse-to-Fine Dual-System
  4. 2026VLA-Forget: Vision-Language-Action Unlearning for Embodied Foundation Models
  5. 2026GigaBrain-0.7: Scaling Embodied Foundation Models to Emergent Capabilities with a Three-System Architecture
  6. 2026ElegantVLA: Learning When to Think for Efficient Vision-Language-Action Models
  7. 2026EmbodiedMidtrain: Bridging the Gap between Vision-Language Models and Vision-Language-Action Models via Mid-training
  8. 2026CoRE-VLA: Towards Scalable and Robust Vision-Language-Action Modeling via Conditional Routing of Experts