Robbyant Open-Sources LingBot-VLA Model for Cross-Platform Robotics

Robbyant, an embodied AI company within Ant Group, has announced in a press release the open-source release of LingBot-VLA, a vision-language-action (VLA) model intended to serve as a universal brain for real-world robots. The model aims to simplify cross-platform deployment and reduce retraining costs associated with robotics development.

LingBot-VLA was evaluated on the GM-100 real-robot benchmark developed by Shanghai Jiao Tong University, where it achieved higher task success rates than other models across three robot platforms. The model also set a new record when depth information was included, improving spatial perception. On the RoboTwin 2.0 simulation benchmark, it demonstrated strong generalization in challenging environments with varied lighting and clutter.

Trained on over 20,000 hours of real-world interaction data from nine dual-arm robot configurations, LingBot-VLA can be adapted to single-arm, dual-arm, and humanoid systems while maintaining performance across different tasks and hardware. The release includes model weights, a production-ready codebase, and tools for data processing, fine-tuning, and automated evaluation.

The announcement was part of Robbyant’s “Evolution of Embodied AI Week” initiative. The company also unveiled LingBot-Depth, a spatial perception model designed to enhance LingBot-VLA’s visual capabilities when used together.