LingBot-VLA 2.0 Open-Sourced and Sped Up

omarsar0 · x · 2026-07-15

LingBot-VLA 2.0 reportedly still faces unbalanced performance in general control; it handles movement and long-horizon scenarios well, but no single policy yet stably covers all operational tasks.

Replies added reproducible details: Robbyant has open-sourced the post-training code; inference with 10 denoising steps takes about 130 ms on a single NVIDIA GeForce RTX 4090D. This allows developers to adapt and test the model without relying on large-scale clusters.

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

Original post →

More from Embodied

Embodied channel →