LightNav-0 Open-Sourced: General Embodied Navigation Model Built on Qwen3-VL

Light Origins has open-sourced LightNav-0, a compact general navigation model that leverages the spatial reasoning of pretrained Qwen3-VL via unified pointing and action tokenization. It achieves state-of-the-art embodied navigation across diverse tasks and robots without task-specific heads.

2026-09-01 ~ 2026-09-01 · 2 related posts

• LightNav-0: Eliciting VLM Spatial Intelligence for Generalist Embodied Navigation — LightOriginsHQ · 2026-09-01
• LightNav-0 Open-Sourced: General-Purpose Robot Navigation Model Using Qwen3-VL — Scobleizer · 2026-09-01