Robbyant/lingbot-vla
A Vision-Language-Action foundation model trained on 20,000 hours of real-world dual-arm robot data for robotic manipulation tasks.

Velocity · 7d
+11
★ / day
Trend
→steady
star history
LingBot-VLA is a pragmatic VLA foundation model for embodied AI and robotics. It trains on large-scale real-world robot data across nine dual-arm configurations, enabling vision-language understanding to drive robotic actions. The project includes pre-training and post-training pipelines, supports LeRobot v3.0, and offers open-loop evaluation with optimized GPU memory usage and Torch Compile support for inference.