Research map / Reinforcement learning / Visual and representation RL
GeoExplorer: Active Geo-localization with Curiosity-Driven Exploration
Li Mi, Manon Béchaz, Zeming Chen, Antoine Bosselut, Devis Tuia
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 state modeling loss (action-state dynamics modeling) (ablation) | GeoExplorer with extrinsic reward · Masa | Success ratio: SR improves 0.0277 on average | Improved |
| add curiosity intrinsic reward without state modeling (ablation) | GeoExplorer · Masa | Success ratio: 0.0357 on average | Improved |
| add learned state modeling to curiosity reward (ablation) | GeoExplorer · Masa | Success ratio: 0.0507 on average | Improved |
| use cosine similarity instead of MSE for intrinsic reward (ablation) | GeoExplorer · Masa | Success ratio: average SR difference of 0.0130 | No clear effect |
| add intrinsic reward (beta 0 to 0.25) (ablation) | GeoExplorer · Masa (C=8) | Success ratio: 0.8648 -> 0.9497 | Improved |
| raise intrinsic weight beta from 0.25 to 0.5 (ablation) | GeoExplorer · Masa (C=4) | Success ratio: 0.4324 -> 0.3597 | Hurt |
| raise intrinsic weight beta from 0.25 to 1 (ablation) | GeoExplorer · Masa (C=8) | Success ratio: 0.9497 -> 0.9352 | Hurt |
| state modeling loss weight alpha 1 versus action loss only (ablation) | GeoExplorer DM stage · Masa action prediction | Accuracy: improvement of 0.0271 over alpha=0 (0.3432 -> 0.3703) | Improved |
| add state modeling and curiosity reward on short paths (ablation) | GeoExplorer · Masa (C=4) | Success ratio: 0.4090 -> 0.3978 | Hurt |
| use GeoExplorer instead of GOMAA-Geo (ablation) | GOMAA-Geo · Masa | Success ratio: 0.0234 at C=4 and 0.1643 at C=8 | Improved |
| use GeoExplorer instead of GOMAA-Geo for unseen targets (ablation) | GOMAA-Geo · SwissViewMonuments | Success ratio: 0.1433 (aerial), 0.1366 (ground) at C=6 | Improved |
| use GeoExplorer instead of GOMAA-Geo in cross-modal goals (ablation) | GOMAA-Geo · MM-GAG | Success ratio: 0.0690 (ground-level) and 0.0610 (text) on average for C={6,7,8} | Improved |
Closest papers by meaning
- GOMAA-Geo: GOal Modality Agnostic Active Geo-localization (NeurIPS 2024 · similarity 0.88)
- Vision-Language Reasoning for Geolocalization: A Reinforcement Learning Approach (AAAI 2026 · similarity 0.87)
- Learning World Models for Unconstrained Goal Navigation (NeurIPS 2024 · similarity 0.83)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 2025 · similarity 0.83)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.83)
- Decoupling Exploration and Policy Optimization: Uncertainty Guided Tree Search for Hard Exploration (NeurIPS 2026 · similarity 0.83)
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.
