Embodied Native Model LingBot-VA 2.0

chris_j_paxton · x · 2026-07-16

The quoted content introduces LingBot-VA 2.0: an embodied native foundation model built from scratch for the physical world, rather than being fine-tuned from a video generator.

It focuses on "thinking ahead and acting in real-time," reporting the following results: a 93.6% success rate on dual-arm tasks, 150 Hz single-GPU inference, and the ability to generalize from just 20 demonstrations. The author emphasizes that the core of this approach is not "adapting" old models, but natively modeling directly for the physical world.

Related event: Ant Group unveils LingBot-VLA 2.0 for cross-embodiment robot control(8 posts)→

Original post →

More from Embodied

Embodied channel →