SimpleVLN: native multimodal LLMs navigate continuously without a specialized backbone
jiqizhixin · x · 2026-11-13
A preprint from Fudan University, NTU and Lenovo Research proposes SimpleVLN, showing native multimodal LLMs can improve continuous navigation via better decisions and training without a specialized navigation backbone. Key choices: (1) each decision takes the instruction, current view and sampled visual history—uniform sampling keeps early route context while recent observations aid local movement—emitting ordered action chunks instead of a growing transcript; (2) training goes expert demos → DAgger corrections → outcome-based RL, exposing states missing from demos and allowing routes that differ from the expert path; (3) results reported with Qwen3-VL-4B.
More from Embodied
- J.P. Morgan forecasts 300k-400k US humanoid robots by 2030, 1.5M by 2035 — xmercury_one · 2026-11-13
- Robopedia launches Xperience 2, a 'gold standard' Physical AI dataset — liuziwei7 · 2026-11-13
- Baidu invites riders to try its 6th-gen Apollo Go robotaxi in Beijing — Baidu_Inc · 2026-11-13
- Gemini Robotics 2.5 Debuts at CoRL 2026: Humanoid Ties a Bow Autonomously — ZoubinGhahrama1 · 2026-11-13
- LLM edits 65 embedding rows to fix robot failures, lifting held-out success 74.3% to 87.5% — 机器之心 · 2026-11-13
- Astra vs Fable 5.2 zero-shot excavator test: only visual reference, one model prevails — Scobleizer · 2026-11-13