Research map / Multi-agent systems and games / Human-AI and goal-conditioned
MAGICIAN: Efficient Long-Term Planning with Imagined Gaussians for Active Mapping
Shiyao Li, Antoine Guédon, Shizhe Chen, Vincent Lepetit
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 |
|---|---|---|---|
| increase beam width and look-ahead steps (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC / final coverage: 6.3% in AUC and 9.3% in final coverage | Improved |
| use Imagined Gaussians coverage gain in greedy mode versus MACARONS Monte Carlo (ablation) | MAGICIAN (beam or depth 1) · Macarons++ 3 unseen scenes (ablation) | AUC / final coverage: 5.2% in AUC and 10.9% in final coverage | Improved |
| compute coverage gain with Imagined Gaussians instead of MACARONS (ablation) | single candidate view · Macarons++ | time per view (s): 0.05s -> 0.002s | Improved |
| fine-tune occupancy model on indoor MP3D (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | final coverage: 0.888 -> 0.893 | No clear effect |
| fine-tune occupancy model on indoor MP3D (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC: 0.652 -> 0.646 | Hurt |
| increase look-ahead steps from 10 to 15 (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC: 0.652 -> 0.673 | Improved |
| increase look-ahead steps from 20 to 50 (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC: 0.664 -> 0.652 | Hurt |
| increase proxy point density from 1x to 4x (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC: 0.652 -> 0.685 | Improved |
| reduce proxy point density from 1x to 0.5x (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | final coverage: 0.888 -> 0.848 | Hurt |
| more frequent replanning (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC / coverage | Improved |
| replan only every 6 steps (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC / coverage | No clear effect |
| add Gaussian pose noise (0.5m, 3 deg) during planning (ablation) | MAGICIAN · Macarons++ 3 unseen scenes (ablation) | AUC: 0.652 -> 0.649 | No clear effect |
Closest papers by meaning
- ActiveGAMER: Active GAussian Mapping through Efficient Rendering (CVPR 2025 · similarity 0.86)
- MAGiC-SLAM: Multi-Agent Gaussian Globally Consistent SLAM (CVPR 2025 · similarity 0.86)
- NextBestPath: Efficient 3D Mapping of Unseen Environments (ICLR 2025 · similarity 0.85)
- Understanding while Exploring: Semantics-driven Active Mapping (NeurIPS 2025 · similarity 0.84)
- Learning with a Mole: Transferable latent spatial representations for navigation without reconstruction (ICLR 2024 · similarity 0.84)
- Open-World Reinforcement Learning over Long Short-Term Imagination (ICLR 2025 · similarity 0.84)
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.
