Research map / Multi-agent systems and games / Human-AI and goal-conditioned
Vision-language navigation agents
Develops and analyzes navigation agents following instructions with structured decision spaces and robustness.
66 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Human-AI and goal-conditioned, keyword cluster “navigation · prediction · embodied”
State History Benchmarks Tried Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 1 of 1 tracked benchmark.
17 papers in 2023–24 and 25 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 5.7% to 6.5%.
Open questions the data points to
- R2R val-unseen (SPL): best went from 62.9 (up to 2024) to 78 in 2025 — Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation (AAAI 2025 · 2 citations)
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · Cognitive Mapping and Planning for Visual Navigation (CVPR 2017 · 537 citations)
- 2018 · Semi-parametric topological memory for navigation (ICLR 2018 · 52 citations)
- 2019 · Self-Monitoring Navigation Agent via Auxiliary Progress Estimation (ICLR 2019 · 185 citations)
- 2020 · Learning to Follow Directions in Street View (AAAI 2020 · 59 citations)
- 2021 · Pushing It Out of the Way: Interactive Visual Navigation (CVPR 2021 · 32 citations)
- 2022 · Bridging the Gap Between Learning in Discrete and Continuous Environments for Vision-and-Language Navigation (CVPR 2022 · 86 citations)
- 2023 · Object-Goal Visual Navigation via Effective Exploration of Relations Among Historical Navigation States (CVPR 2023 · 36 citations)
- 2024 · VELMA: Verbalization Embodiment of LLM Agents for Vision and Language Navigation in Street View (AAAI 2024 · 61 citations)
- 2025 · Navigation World Models (CVPR 2025 · 34 citations)
- 2026 · SocialNav: Training Human-Inspired Foundation Model for Socially-Aware Embodied Navigation (CVPR 2026 · 1 citation)
Earliest papers here
- Cognitive Mapping and Planning for Visual Navigation (CVPR 2017 · 537 citations)
- Learning to Navigate in Complex Environments (ICLR 2017 · 356 citations)
- Semi-parametric topological memory for navigation (ICLR 2018 · 52 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
R2R val-unseen · SPL (higher is better)
4 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 62.9 | EnvEdit: Environment Editing for Vision-and-Language Navigation (CVPR 2022 · 81 citations) |
| 2025 | 78 | Planning from Imagination: Episodic Simulation and Episodic Memory for Vision-and-Language Navigation (AAAI 2025 · 2 citations) |
What has been tried
51 papers here report 539 measured changes (a component added, removed or swapped, compared with a baseline): helped 65%, no clear effect 9%, hurt 24%, mixed 3%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Train with response-based learning instead of teacher-forcing alone | Touchdown development · TC | +2.9 | Helped | VELMA: Verbalization Embodiment of LLM Agents for Vision and Language Navigation in Street View (AAAI 2024 · 61 citations) |
| Use 4 goals per sample instead of 1 | RECON, 4 seconds ahead · LPIPS | 0.312 ± 0.001 -> 0.296 ±0.002 | Helped | Navigation World Models (CVPR 2025 · 34 citations) |
| add intra-loop closure | Replica · global reconstruction accuracy (cm) (↓) | 1.831 -> 1.681 | Helped | MNE-SLAM: Multi-Agent Neural SLAM for Mobile Robots (CVPR 2025 · 28 citations) |
| Add active camera movement instead of fixing the camera forward | Simulation cluttered navigation · success rate | ≈68% | Helped | SPIN: Simultaneous Perception Interaction and Navigation (CVPR 2024 · 14 citations) |
| add loop closure | ReplicaMultiagent · ATE RMSE [cm] | 0.42 -> 0.27 (Off-0 average); 0.38 -> 0.16 (Apt-0 average); 0.54 -> 0.26 (Apt-1 average); 0.66 -> 0.32 (Apt-2 average) | Helped | MAGiC-SLAM: Multi-Agent Gaussian Globally Consistent SLAM (CVPR 2025 · 12 citations) |
| Add orientation loss | Offline teleoperation test, scenario mean · MAOE (°) | 17.03 -> 17.00 | No clear effect | CityWalker: Learning Embodied Urban Navigation from Web-Scale Videos (CVPR 2025 · 7 citations) |
| Add the forgetting module that excludes low-attention nodes | Gibson 1-goal · SR | 52.1 -> 55.1 | Helped | MemoNav: Working Memory Model for Visual Navigation (CVPR 2024 · 7 citations) |
| add identified second-order dynamical motion model in simulation training (vs. constant-velocity sim) | Sim (+dyn.) HM3D/2.5k · SR% | 0.8 -> 97.4 | Helped | Learning to Navigate Efficiently and Precisely in Real Environments (CVPR 2024 · 6 citations) |
| add screenshots to HTML inputs | MiniWoB++ · success rate | 55.7% -> 61.1% | Helped | Multimodal Web Navigation with Instruction-Finetuned Foundation Models (ICLR 2024 · 6 citations) |
| set δques to 0.0 rather than 0.5 | Unheard Sound · Success | 36.5 -> 42.0 | Helped | CAVEN: An Embodied Conversational Agent for Efficient Audio-Visual Navigation in Noisy Environments (AAAI 2024 · 4 citations) |
| add sequence trajectory training (length 32) instead of a single waypoint (length 1, i-th = 1) | Gibson val · SR (%) | 71.0 -> 79.6 | Helped | Trajectory Diffusion for ObjectGoal Navigation (NeurIPS 2024 · 4 citations) |
| zero latent memory periodically | Test-bldg/14 · SR | 100.0 -> 85.0 (30 sec); 100.0 -> 80.0 (10 sec); 100.0 -> 75.0 (3 sec) | Hurt | Reasoning in Visual Navigation of End-to-end Trained Agents: A Dynamical Systems Approach (CVPR 2025 · 3 citations) |
What reviewers will expect
Datasets and baselines named in the 15 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- HM3D · 4 of 15 papers
- R2R · 2 of 15 papers
- REVERIE · 2 of 15 papers
- Gibson · 2 of 15 papers
- MP3D · 2 of 15 papers
- R2R-CE · 2 of 15 papers
Baselines
- BEVBert · 3 of 15 papers
- DUET · 2 of 15 papers
- ZSON · 2 of 15 papers
- GNM · 2 of 15 papers
- NoMaD · 2 of 15 papers
- CP-SLAM · 2 of 15 papers
- CCM-SLAM · 2 of 15 papers
- Swarm-SLAM · 2 of 15 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 42 papers; from OpenAlex and the papers' affiliations).
- University of California, Santa Barbara · 3 papers
- Xi’an Jiaotong University · 3 papers
- Samsung · 2 papers
- Shanghai Jiao Tong University · 2 papers
- Georgia Institute of Technology · 2 papers
- University of Chinese Academy of Sciences · 2 papers
- NAVER · 2 papers
- New York University · 2 papers
- Chinese Academy of Sciences · 2 papers
- AIMdyn Inc · 2 papers
Use it
Download BibTeX of 37 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Embodied agent learning (93 papers)
- Interactive multi-agent trajectory prediction (75 papers)
- Data-driven traffic simulation (36 papers)
Elsewhere in the topic
- Human-AI coordination agents (125 papers)
- World-model-based agents (110 papers)
- Imitation and skill learning (100 papers)
- Language-grounded reinforcement agents (96 papers)
- Boundedly rational agent modeling (87 papers)
- Collaborative multi-agent perception (86 papers)
- Intrinsically motivated exploration (86 papers)
- Learning-guided MAPF search (73 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).