Robbyant, an AI company within Ant Group in Shanghai, has open-sourced LingBot-VLA, a vision-language-action model designed as a universal control system for robots.

The model has been tested on robots from Galaxea Dynamics and AgileX Robotics, showing strong cross-platform transfer capabilities.

On the GM-100 benchmark of 100 real-world tasks, LingBot-VLA outperformed other models, especially when combined with depth perception.

LingBot-VLA also performed well on the RoboTwin 2.0 simulation benchmark, handling complex environments and variations.

It was trained on more than 20,000 hours of real-world data across nine robot configurations, reducing the need for retraining on new hardware.

The open-source release includes model weights and a full codebase, with CEO Zhu Xing saying the goal is to make embodied AI more scalable and cost-effective.

🔗 Source: Robbyant