[📜 Paper] [⭐️Project Page] [🤗 Model] [🤗 Dataset]

While significant research has focused on developing embodied reasoning capabilities using Vision-Language Models (VLMs) or integrating advanced VLMs into Vision-Language-Action (VLA) models for end-to-end robot control, few studies directly address the critical gap between upstream VLM-based reasoning and downstream VLA policy learning. In this work, we take an initial step toward bridging embodied reasoning with VLA policy learning by introducing Vlaser -- a Vision-Language-Action Model with synergistic embodied reasoning capability, which is a foundational vision-language model designed to integrate high-level reasoning with low-level control for embodied agents. Built upon the high-quality Vlaser-6M dataset, Vlaser achieves state-of-the-art performance across a range of embodied reasoning benchmarks—including spatial reasoning, embodied grounding, embodied QA, and task planning. Furthermore, we systematically examine how different VLM initializations affect supervised VLA fine-tuning, offering novel insights into mitigating the domain shift between internet-scale pre-training data and embodied-specific policy learning data. Based on these insights, our approach achieves state-of-the-art results on the WidowX benchmark and competitive performance on the Google Robot benchmark.

• 2025-10-13: 🤖 We release Vlaser VLM model (Vlaser-2B and Vlaser-8B) as well as VLA model (Vlaser-2B-VLA) on 🤗Vlaser.
• 2025-10-13: 🤖 We release the training and inference code of Vlaser VLM based on InternVL3.
• 2025-11-7: 🤖 We release the training and inference code of Vlaser VLA based on open-pi-zero.
• 2026-01-27: 🤖 Vlaser was accepted by ICLR 2026, congrats!
• 2026-02-15: 🤖 We release the data pipeline for in-domain data based on open-pi-zero.
• 2026-03-18: 🤖 We release the training dataset 🤗Vlaser-6M, which could help you train your own embodied brain! 🔥🔥🔥

• Release Vlaser-2B and Vlaser-8B ckpt for VLM embodied reasoning.
• Release Vlaser-2B-VLA model for end-to-end robot control in SimplerEnv (WidowX and Google Robot) and RoboTwin 2.0.
• Release the training and evaluation code for Vlaser VLMs.
• Release the training and evaluation code for Vlaser VLAs.
• Release the Dataset Generation Pipeline.
• Release the Vlaser-6M Dataset.

Please refer to Vlaser_VLM for details.

For SimplerEnv, Please refer to Vlaser_VLA/Simpler for details. For RoboTwin 2.0, Please refer to Vlaser_VLA/RoboTwin for details.

Please refer to data-pipeline for details.

You can download our training dataset 🤗Vlaser-6M, which is categorized into Robot_QA_data (for general Robot QA tasks), grounding_data (for 2D robot grounding tasks), planning_data (for robotic planning tasks) and spatial_data (for spatial intelligence tasks). Each dataset is composed of a *.jsonl file for multimodal annotation and a *.tar.gz file for images/videos.

This project is released under the MIT License.

If you find this work helpful in your research, please consider giving this repo a star ⭐ and citing our paper:

@article{yang2025vlaser,
  title={Vlaser: Vision-Language-Action Model with Synergistic Embodied Reasoning},
  author={Yang, Ganlin and Zhang, Tianyi and Hao, Haoran and Wang, Weiyun and Liu, Yibin and Wang, Dehui and Chen, Guanzhou and Cai, Zijian and Chen, Junting and Su, Weijie and others},
  journal={arXiv preprint arXiv:2510.11027},
  year={2025}
}