DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Action

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Fang, Zhen, Liu, Zhuoyang, Liu, Jiaming, Chen, Hao, Zeng, Yu, Huang, Shiting, Chen, Zehui, Chen, Lin, Zhang, Shanghang, Zhao, Feng
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866909928807989248
author Fang, Zhen
Liu, Zhuoyang
Liu, Jiaming
Chen, Hao
Zeng, Yu
Huang, Shiting
Chen, Zehui
Chen, Lin
Zhang, Shanghang
Zhao, Feng
author_facet Fang, Zhen
Liu, Zhuoyang
Liu, Jiaming
Chen, Hao
Zeng, Yu
Huang, Shiting
Chen, Zehui
Chen, Lin
Zhang, Shanghang
Zhao, Feng
contents To build a generalizable Vision-Language-Action (VLA) model with strong reasoning ability, a common strategy is to first train a specialist VLA on robot demonstrations to acquire reliable manipulation skills, and then incorporate mixed annotated robot data together with multimodal data to restore broader reasoning capabilities. However, we observe that the resulting reasoning VLA often suffers from degraded action performance compared to the specialist model before fine-tuning, a phenomenon we refer to as action degeneration. To address this issue, we propose DualVLA, which enhances action performance through carefully designed post-training while still preserving reasoning capability. We first introduce a dual-layer data pruning method that removes redundant embodied reasoning, preventing it from adversely influencing action learning. To further strengthen action generation, we design a dual-teacher adaptive distillation strategy that assigns different supervision signals to different data domains while maintaining reasoning ability. To fill the evaluation gap for generalist VLAs, we also propose VLA Score, which decouples VLA capability into reasoning, intention, action, and alignment dimensions for a more fine-grained assessment. Experiments show that DualVLA achieves an average success rate of 61.0 in SimplerEnv and an average score of 65.4 across eight competitive multimodal benchmarks, demonstrating a stronger balance between precise action execution and multimodal understanding. Project Website: https://costaliya.github.io/DualVLA/.
format Preprint
id arxiv_https___arxiv_org_abs_2511_22134
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Action
Fang, Zhen
Liu, Zhuoyang
Liu, Jiaming
Chen, Hao
Zeng, Yu
Huang, Shiting
Chen, Zehui
Chen, Lin
Zhang, Shanghang
Zhao, Feng
Computer Vision and Pattern Recognition
Robotics
To build a generalizable Vision-Language-Action (VLA) model with strong reasoning ability, a common strategy is to first train a specialist VLA on robot demonstrations to acquire reliable manipulation skills, and then incorporate mixed annotated robot data together with multimodal data to restore broader reasoning capabilities. However, we observe that the resulting reasoning VLA often suffers from degraded action performance compared to the specialist model before fine-tuning, a phenomenon we refer to as action degeneration. To address this issue, we propose DualVLA, which enhances action performance through carefully designed post-training while still preserving reasoning capability. We first introduce a dual-layer data pruning method that removes redundant embodied reasoning, preventing it from adversely influencing action learning. To further strengthen action generation, we design a dual-teacher adaptive distillation strategy that assigns different supervision signals to different data domains while maintaining reasoning ability. To fill the evaluation gap for generalist VLAs, we also propose VLA Score, which decouples VLA capability into reasoning, intention, action, and alignment dimensions for a more fine-grained assessment. Experiments show that DualVLA achieves an average success rate of 61.0 in SimplerEnv and an average score of 65.4 across eight competitive multimodal benchmarks, demonstrating a stronger balance between precise action execution and multimodal understanding. Project Website: https://costaliya.github.io/DualVLA/.
title DualVLA: Building a Generalizable Embodied Agent via Partial Decoupling of Reasoning and Action
topic Computer Vision and Pattern Recognition
Robotics
url https://arxiv.org/abs/2511.22134