Research map / 3D vision / Gaussian splatting
EmbodiedSplat: Personalized Real-to-Sim-to-Real Navigation with Gaussian Splats from a Mobile Device
Gunjan Chhablani, Xiaomeng Ye, Muhammad Zubair Irshad, Zsolt Kira
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 |
|---|---|---|---|
| fine-tune on Polycam mesh of deployment scene | HM3D pre-trained policy · Real-world lounge ImageNav | success rate: 50% -> 70% | Improved |
| fine-tune on DN mesh of deployment scene | HM3D pre-trained policy · Real-world lounge ImageNav | success rate: 50% -> 70% | Improved |
| fine-tune on Polycam mesh | HSSD pre-trained policy · Real-world lounge ImageNav | success rate: 10% -> 50% | Improved |
| fine-tune on DN mesh | HSSD pre-trained policy · Real-world lounge ImageNav | success rate: 10% -> 40% | Improved |
| fine-tune 20M steps on scene mesh | HM3D pre-trained policy · Sim Captured/MuSHRoom scenes | success rate: approaching 90%+ | Improved |
| fine-tune 20M steps on scene mesh | HM3D pre-trained policy · conf_c (DN mesh) | success rate: 0.16 -> 0.84 | Improved |
| add HM3D/HSSD large-scale pre-training before scene training (ablation) | policy · Real-world lounge ImageNav (DN mesh) | success rate: 10% -> 70% | Improved |
| add HM3D/HSSD pre-training before scene training (ablation) | policy · Real-world lounge ImageNav (Polycam mesh) | success rate: 50% -> 70% | Improved |
| continue HM3D pre-training beyond ~400M steps (ablation) | HM3D policy · Captured scenes zero-shot | success rate | No clear effect |
| continue HSSD pre-training beyond 300M steps (ablation) | HSSD policy · Captured scenes zero-shot | success rate | No clear effect |
| use iPhone GT depth instead of monocular depth encoders (ablation) | DN-Splatter · MuSHRoom meshes | F-score: 0.688 -> 0.748 | Improved |
| use Metric3D-v2 normals instead of Omnidata (ablation) | DN-Splatter · MuSHRoom meshes | F-score: 0.748 -> 0.752 | Improved |
Closest papers by meaning
- Habitat-GS: A High-Fidelity Navigation Simulator with Dynamic Gaussian Splatting (ECCV 2026 · similarity 0.91)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.89)
- Habitat: A Platform for Embodied AI Research (ICCV 2019 · similarity 0.89)
- Habitat Synthetic Scenes Dataset (HSSD-200): An Analysis of 3D Scene Scale and Realism Tradeoffs for ObjectGoal Navigation (CVPR 2024 · similarity 0.88)
- Learning to Navigate Efficiently and Precisely in Real Environments (CVPR 2024 · similarity 0.87)
- InternScenes: A Large-scale Simulatable Indoor Scene Dataset with Realistic Layouts (NeurIPS 2025 · similarity 0.86)
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.
