Research map / Vision-language and multimodal / Robot action and navigation
OctoNav: Towards Generalist Embodied Navigation
Chen Gao, Liankai Jin, Xingyu Peng, Jiazhao Zhang, Yue Deng, Annan Li, He Wang, Si Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add Action-SFT on OctoNav-Bench pairs (ablation) | Base-Model (NaVid) · OctoNav-Bench test (Overall) | SR (%): 5.80 -> 8.80 | Improved |
| add Action-SFT on OctoNav-Bench pairs (ablation) | Base-Model (NaVid) · OctoNav-Bench test (VLN) | VLN SR (%): 25.71 -> 20.00 | Hurt |
| add TBA-SFT (think-before-action) (ablation) | Action-SFT model · OctoNav-Bench test (Overall) | SR (%): 8.80 -> 14.40 | Improved |
| add TBA-SFT (think-before-action) (ablation) | Action-SFT model · OctoNav-Bench test (PointNav) | PointNav SR (%): +11.55% | Improved |
| add Nav-GRPO (ablation) | TBA-SFT model · OctoNav-Bench test (Overall) | SR (%): 14.40 -> 17.00 | Improved |
| add online RL (ablation) | Nav-GRPO model · OctoNav-Bench test (Overall) | SR (%): 17.00 -> 19.40 | Improved |
| add online RL (ablation) | Nav-GRPO model · OctoNav-Bench test (VLN) | VLN SR (%): +2.85% | Improved |
| use stepped reward instead of strict reward (ablation) | Nav-GRPO · OctoNav-Bench test (Overall) | SR (%): 16.20 -> 17.00 | Improved |
| use stepped reward instead of loose reward (ablation) | Nav-GRPO · OctoNav-Bench test (Overall) | SR (%): 15.40 -> 17.00 | Improved |
| use diverse prompt templates instead of a single one (ablation) | Nav-GRPO · OctoNav-Bench test (Overall) | SR (%): 15.80 -> 17.00 | Improved |
| think every 20 steps instead of every 10 (ablation) | OctoNav-R1 · OctoNav-Bench test (Overall) | SR (%): 17.00 -> 19.40 | Improved |
| think every 20 steps instead of every 40 (ablation) | OctoNav-R1 · OctoNav-Bench test (Overall) | SR (%): 19.40 vs 18.80 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- NavGPT-2* — NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Towards Learning a Generalist Model for Embodied Navigation (CVPR 2024 · similarity 0.88)
- NavGPT: Explicit Reasoning in Vision-and-Language Navigation with Large Language Models (AAAI 2024 · similarity 0.88)
- RAO-Nav: Probing Omni-Language Models for Zero-shot Semantic Audio-Visual Navigation (NeurIPS 2026 · similarity 0.88)
- NavBench: Probing Multimodal Large Language Models for Embodied Navigation (NeurIPS 2025 · similarity 0.88)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.88)
- Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
