LightNav-0 Open-Sourced: General-Purpose Robot Navigation Model Using Qwen3-VL

Scobleizer · x · 2026-09-01

Light Origins has released LightNav-0, an open-source general-purpose embodied navigation model. It elicits the spatial intelligence of a pretrained vision-language model (Qwen3-VL) and aligns it with navigation tasks without task-specific prediction heads. Trained entirely in simulation, it supports humanoid, quadruped, wheeled, and aerial robots, enabling scalable real2sim2real transfer across robots, tasks, and scenes. The project includes code, model weights, a paper, and a technical blog.

Related event: LightNav-0 Open-Sourced: General Embodied Navigation Model Built on Qwen3-VL(2 posts)→

Original post →

More from Embodied

Embodied channel →