Research map / 3D vision / 3D generation
GLEAM: Learning Generalizable Exploration Policy for Active Mapping in Complex 3D Indoor Scene
Xiao Chen, Tai Wang, Quanyi Li, Tao Huang, Jiangmiao Pang, Tianfan Xue
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 |
|---|---|---|---|
| use long-term goal action space instead of short-term goal (ablation) | GLEAM · 128 unseen scenes | Coverage: 48.85% -> 66.50% | Improved |
| use random initial poses instead of fixed initialization (ablation) | GLEAM · 128 unseen scenes | Coverage: 56.05% -> 66.50% | Improved |
| use semantic occupancy map instead of binary occupancy map (ablation) | GLEAM · 128 unseen scenes | Coverage: 62.92% -> 66.50% | Improved |
| use semantic occupancy map instead of probabilistic occupancy map (ablation) | GLEAM · 128 unseen scenes | Coverage: 64.37% -> 66.50% | Improved |
| use LocoTransformer encoder instead of UNet (ablation) | GLEAM · 128 unseen scenes | Coverage: 64.01% -> 66.50% | Improved |
| update training scenes with probability 1 instead of 0.05 (ablation) | GLEAM · 128 unseen scenes | Coverage: 63.17% -> 66.50% | Improved |
| add Gibson and HSSD data to ProcTHOR-only training (ablation) | GLEAM · 128 unseen scenes | Coverage: 65.50% -> 66.50% | Improved |
| train on 192 scenes with at least 10 rooms versus 416 simple scenes (ablation) | GLEAM · 128 unseen scenes | Coverage: 61.66% -> 63.23% | Improved |
| scale training scenes from 96 Gibson to 1024 mixed scenes (ablation) | GLEAM · 128 unseen scenes | Coverage: 50.32% -> 66.50% | Improved |
| add a generic path-efficiency reward (ablation) | GLEAM (128 scenes, 2.5k iterations) · 128 unseen scenes | Coverage: 60.23% -> 56.61% | Hurt |
| add cumulative pose noise (variance 0.5) at inference | GLEAM · 128 unseen scenes | Coverage: 66.50% -> 55.59% | Hurt |
| add depth noise (variance 0.2) at inference | GLEAM · 128 unseen scenes | Coverage: 66.50% -> 54.77% | Hurt |
Closest papers by meaning
- Understanding while Exploring: Semantics-driven Active Mapping (NeurIPS 2025 · similarity 0.88)
- NextBestPath: Efficient 3D Mapping of Unseen Environments (ICLR 2025 · similarity 0.88)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · similarity 0.85)
- Habitat Synthetic Scenes Dataset (HSSD-200): An Analysis of 3D Scene Scale and Realism Tradeoffs for ObjectGoal Navigation (CVPR 2024 · similarity 0.85)
- MNE-SLAM: Multi-Agent Neural SLAM for Mobile Robots (CVPR 2025 · similarity 0.85)
- GenNBV: Generalizable Next-Best-View Policy for Active 3D Reconstruction (CVPR 2024 · similarity 0.85)
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.
