Enregistré dans:
Détails bibliographiques
Auteurs principaux: Yu, Bin, Lian, Shijie, Lin, Xiaopeng, Wei, Yuliang, Shen, Zhaolong, Wu, Changti, Miao, Yuzhuo, Wang, Xinming, Wang, Bailing, Huang, Cong, Chen, Kai
Format: Preprint
Publié: 2026
Sujets:
Accès en ligne:https://arxiv.org/abs/2601.14133
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866914294525853696
author Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Wei, Yuliang
Shen, Zhaolong
Wu, Changti
Miao, Yuzhuo
Wang, Xinming
Wang, Bailing
Huang, Cong
Chen, Kai
author_facet Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Wei, Yuliang
Shen, Zhaolong
Wu, Changti
Miao, Yuzhuo
Wang, Xinming
Wang, Bailing
Huang, Cong
Chen, Kai
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2601_14133
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers
Yu, Bin
Lian, Shijie
Lin, Xiaopeng
Wei, Yuliang
Shen, Zhaolong
Wu, Changti
Miao, Yuzhuo
Wang, Xinming
Wang, Bailing
Huang, Cong
Chen, Kai
Robotics
Computer Vision and Pattern Recognition
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.
title TwinBrainVLA: Unleashing the Potential of Generalist VLMs for Embodied Tasks via Asymmetric Mixture-of-Transformers
topic Robotics
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2601.14133