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

Bin Yu^1,2,* Shijie Lian^2,4,* Xiaopeng Lin^2,5,* Yuliang Wei^1,† Zhaolong Shen^2,6
Changti Wu^2,7 Yuzhuo Miao^1,2 Xinming Wang^2,8 Bailing Wang¹ Cong Huang^2,3 Kai Chen^2,3,9,†

¹HIT, ²ZGCA, ³ZGCI, ⁴HUST, ⁵HKUST(GZ), ⁶BUAA, ⁷ECNU, ⁸CASIA, ⁹DeepCybo

^*Equal contribution, ^†Corresponding author

Zhongguancun Academy & Zhongguancun Institute of Artificial Intelligence

Standard Vision-Language-Action (VLA) models typically fine-tune a monolithic Vision-Language Model (VLM) backbone explicitly for robotic control. However, this approach creates a critical tension between maintaining high-level general semantic understanding and learning low-level, fine-grained sensorimotor skills, often leading to "catastrophic forgetting" of the model's open-world capabilities.

To resolve this conflict, we introduce TwinBrainVLA, a novel architecture that coordinates a generalist VLM retaining universal semantic understanding and a specialist VLM dedicated to embodied proprioception for joint robotic control. TwinBrainVLA synergizes a frozen "Left Brain", which retains robust general visual reasoning, with a trainable "Right Brain", specialized for embodied perception, via a novel Asymmetric Mixture-of-Transformers (AsyMoT) mechanism. This design allows the Right Brain to dynamically query semantic knowledge from the frozen Left Brain and fuse it with proprioceptive states, providing rich conditioning for a Flow-Matching Action Expert to generate precise continuous controls.

Extensive experiments on SimplerEnv and RoboCasa benchmarks demonstrate that TwinBrainVLA achieves superior manipulation performance compared to state-of-the-art baselines while explicitly preserving the comprehensive visual understanding capabilities of the pre-trained VLM, offering a promising direction for building general-purpose robots that simultaneously achieve high-level semantic understanding and low-level physical dexterity.

TwinBrainVLA mimics the biological principle of hemispheric lateralization:

• Left Brain (Generalist): A frozen, pre-trained VLM (e.g., Qwen-VL) that serves as a semantic anchor, preserving open-world knowledge.
• Right Brain (Specialist): A trainable VLM initialized with the same weights, specialized for embodied control and proprioceptive state encoding.
• Asymmetric MoT (AsyMoT): A mechanism where the Right Brain attends to the frozen Key-Value (KV) pairs of the Left Brain via causal self-attention, transferring semantic knowledge without parameter pollution.
• Action Expert: A Flow-Matching Diffusion Transformer (DiT) that generates continuous actions based on the condition features from the Right Brain.

We would like to thank the starVLA project for its inspiring work and open-source contributions. At the same time, we also express our gratitude to the following projects:

• Mixture-of-Transformers
• Isaac-GR00T
• LeRobot
• SimplerEnv
• Franka Teleop

If you find this project or the dataset helpful, please cite:

@misc{TwinBrainVLA,
      title={TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers}, 
      author={Bin Yu and Shijie Lian and Xiaopeng Lin and Yuliang Wei and Zhaolong Shen and Changti Wu and Yuzhuo Miao and Xinming Wang and Bailing Wang and Cong Huang and Kai Chen},
      year={2026},
      eprint={2601.14133},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2601.14133}, 
}