Robbyant releases LingBot-VLA 2.0, an open-source VLA model for cross-embodiment robotics

The 6B-parameter VLA model is trained on 50k hours real robot data plus 10k hours human video, unifying 20 robot bodies. It beats π0.5 and GR00T N1.7 on the GM-100 benchmark and runs at ~130 ms on an RTX 4090D.