Research map / 3D vision / Gaussian splatting
Towards Physically Executable 3D Gaussian for Embodied Navigation
Bingchen Miao, Rong Wei, Zhiqi Ge, Xiaoquan sun, Shiqi Gao, Jingzhe Zhu, Renhan Wang, Siliang Tang, Jun Xiao, Rui Tang, Juncheng Li
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 NaVILA-base on 500k SAGE-Bench trajectory-instruction pairs (3DGS data) | NaVILA · VLN-CE R2R Val-Unseen | SR: 0.29 -> 0.38 | Improved |
| fine-tune NaVid-base on SAGE-Bench data | NaVid · VLN-CE R2R Val-Unseen | SR: 0.22 -> 0.31 | Improved |
| fine-tune NaVILA-base on SAGE-Bench data | NaVILA · SAGE-Bench VLN (high-level) | SR: 0.21 -> 0.46 | Improved |
| fine-tune NaVid-base on SAGE-Bench data | NaVid · SAGE-Bench VLN (high-level) | SR: 0.10 -> 0.36 | Improved |
| use 3DGS-Mesh hybrid environment instead of scanned mesh (ablation) | NaVILA-base · rendering speed | render time per frame (ms): 16.7 -> 6.2 | Improved |
| use 3DGS data instead of scanned mesh for training (ablation) | NaVILA-base · training convergence | time to SR=40% (hrs): 4.8 -> 6.2 | Hurt |
| increase number of training scenes at fixed sample size vs increase samples (ablation) | SAGE-trained VLN model · VLN-CE val-unseen | performance | Improved |
| increase sampling density at a fixed number of scenes (800) (ablation) | SAGE-trained VLN model · VLN-CE val-unseen | performance | Improved |
| fine-tune NaVILA on SAGE data (continuity trade-off) | NaVILA-SAGE · SAGE-Bench VLN | ICP and PS vs original NaVILA: original NaVILA ICP 0.61, PS 0.68; NaVILA-SAGE ICP 0.54, PS 0.74 | Mixed |
| low-level vs high-level instructions | NaVILA · SAGE-Bench VLN | SR: 0.56 -> 0.39 | Hurt |
| longer trajectories and higher scene complexity | VLN models · SAGE-Bench slices | SR | Hurt |
Closest papers by meaning
- GaussNav: Gaussian Splatting for Visual Navigation (TPAMI 2025 · similarity 0.90)
- SceneSplat++: A Large Dataset and Comprehensive Benchmark for Language Gaussian Splatting (NeurIPS 2025 · similarity 0.89)
- PlanarGS: High-Fidelity Indoor 3D Gaussian Splatting Guided by Vision-Language Planar Priors (NeurIPS 2025 · similarity 0.87)
- Image2Sim: Scaling Embodied Navigation via Generative Neural Simulator (NeurIPS 2026 · similarity 0.87)
- FastLGS: Speeding Up Language Embedded Gaussians with Feature Grid Mapping (AAAI 2025 · similarity 0.87)
- Tackling View-Dependent Semantics in 3D Language Gaussian Splatting (ICML 2025 · similarity 0.87)
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.
