CombatVLA: An Efficient Vision-Language-Action Model for Combat Tasks in 3D Action Role-Playing Games

Recent advances in Vision-Language-Action models (VLAs) have expanded the capabilities of embodied intelligence. However, significant challenges remain in real-time decision-making in complex 3D environments, which demand second-level responses, high-resolution perception, and tactical reasoning under dynamic conditions. To advance the field, we introduce CombatVLA, an efficient VLA model optimized for combat tasks in 3D action role-playing games(ARPGs). Specifically, our CombatVLA is a 3B model trained on video-action pairs collected by an action tracker, where the data is formatted as action-of-thought (AoT) sequences. Thereafter, CombatVLA seamlessly integrates into an action execution framework, allowing efficient inference through our truncated AoT strategy. Experimental results demonstrate that CombatVLA not only outperforms all existing models on the combat understanding benchmark but also achieves a 50-fold acceleration in game combat. Moreover, it has a higher task success rate than human players. We will open-source all resources, including the action tracker, dataset, benchmark, model weights, training code, and the implementation of the framework atthis https URL.

@article{chen2025_2503.09527,
  title={ CombatVLA: An Efficient Vision-Language-Action Model for Combat Tasks in 3D Action Role-Playing Games },
  author={ Peng Chen and Pi Bu and Yingyao Wang and Xinyi Wang and Ziming Wang and Jie Guo and Yingxiu Zhao and Qi Zhu and Jun Song and Siran Yang and Jiamang Wang and Bo Zheng },
  journal={arXiv preprint arXiv:2503.09527},
  year={ 2025 }
}