# Two Bridges, One Pathway: From VLMs to Generalizable VLAs with Embodied Trajectory-Coupled Data

Source: [Two Bridges, One Pathway: From VLMs to Generalizable VLAs with Embodied Trajectory-Coupled Data](<https://www.sota2.com/research/paper/6a279e2090966ade919bf75d>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-06-10T05:01:35.776Z

Authors: Linqi Yin, Shiduo Zhang, Shenling Qiu, Chenxin Li, Zhaoyang Fu, Lei Xiao, Xiang Wang, Chenchen Yang, Zhe Xu, Pengfang Qian, Jingjing Gong, Xipeng Qiu, Xuanjing Huang, Yu-Gang Jiang

Published: 2026-06-07T08:57:51.000Z

## Abstract

Vision-language models (VLMs) are powerful general-purpose reasoners, yet converting them into robot control policies (VLAs) is surprisingly difficult. The root cause is a two-fold gap: VLMs are trained on internet-scale images with language-understanding objectives, while VLAs must perceive robot scenes and predict motor actions. Fine-tuning a VLM directly on robot action data forces the model to cross both gaps at once -- the learning curve is steep and the rich generalizations learned during pretraining tend to degrade rather than transfer. We argue that this gap can be bridged gradually with the right intermediate data. We introduce \\emph{embodied trajectory-coupled (ETC) data} -- vision-language supervision derived from the same robot scenes and trajectories used for action learning. Because ETC data shares the visual context of robot operation while retaining familiar language-understanding objectives, it provides a natural stepping stone between VLM pretraining and VLA fine-tuning. Building on this, we design a three-stage training recipe. Distribution Bridging first adapts the VLM to embodied visual-language semantics. Objective Bridging then gradually shifts the model toward action prediction while preserving the acquired representations. Retentive Adaptation finally specializes the policy to the target deployment domain. We further show that mixing task-relevant out-of-distribution ETC data with a small amount of action data enables the model to generalize to novel visual-language conditions without requiring additional robot demonstrations. Simulation and real-robot experiments confirm that this gradual bridging strategy is the key to transferring VLM generalization into robust, deployable robot policies.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Robot Action Prediction on SimplerEnv](<https://www.sota2.com/research/sota/robot-action-prediction-on-simplerenv.md>): Robot Action Prediction; SimplerEnv; Success Rate: 84.4.

- [Robot Action Prediction on VLABench T1](<https://www.sota2.com/research/sota/robot-action-prediction-on-vlabench-t1.md>): Robot Action Prediction; VLABench T1; Success Rate (SR): 64.8.

- [Robot Action Prediction on VLABench T2](<https://www.sota2.com/research/sota/robot-action-prediction-on-vlabench-t2.md>): Robot Action Prediction; VLABench T2; Success Rate: 39.9.

- [Robot Task Completion on LIBERO (Success Rate and Wall-Clock Time)](<https://www.sota2.com/research/sota/robot-task-completion-on-libero-success-rate-and-wall-clock-time.md>): Robot Task Completion; LIBERO; Success Rate (Mean): 90.7.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2606.08520>)

[PDF](<https://arxiv.org/pdf/2606.08520.pdf>)