Research map / 3D vision / 3D generation
SIMSplat: Language-Aligned 4D Gaussian Splatting for Driving Scenario Generation
Sung-Yeon Park, Adam Lee, Juanwu Lu, Can Cui, Luyang Jiang, Rohit Gupta, Kyungtae Han, Ahmadreza Moradipari, Ziran Wang
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 multi-agent path refinement (ablation) | SIMSplat · Motion generation | failure rate (%): 53.3 -> 20.0 | Improved |
| add multi-agent path refinement (ablation) | SIMSplat · Motion generation | vehicle collision rate (%): 43.3 -> 13.3 | Improved |
| add multi-agent path refinement (ablation) | SIMSplat · Motion generation | pedestrian collision rate (%): 26.7 -> 16.7 | Improved |
| add multi-agent path refinement (ablation) | SIMSplat · Motion generation | off-road rate (%): 10.0 -> 6.7 | Improved |
| add motion and location features to appearance features (ablation) | SIMSplat language alignment · Road object querying | accuracy and vIoU | Improved |
| add motion features to appearance features (ablation) | SIMSplat language alignment · Road object querying (vehicles) | accuracy and vIoU | Improved |
| add location features to appearance features (ablation) | SIMSplat language alignment · Road object querying (pedestrians) | accuracy and vIoU | Improved |
| use motion-aware alignment instead of 4DLangSplat | language-aligned Gaussians · Road object querying (vehicles) | accuracy: 0.83 accuracy, nearly doubling 4DLangSplat | Improved |
| use motion-aware alignment instead of 4DLangSplat | language-aligned Gaussians · Road object querying (pedestrians) | accuracy and vIoU: 70.3% and 77.3% higher | Improved |
| replace Qwen3-VL-8B with Qwen3-VL-32B (ablation) | VLM-driven scenario mining · Task completion | task completion (%): 48.3 -> 71.1 | Improved |
| replace Qwen3-VL-32B with GPT-5 (ablation) | VLM-driven scenario mining · Task completion | task completion (%): 71.1 -> 78.9 | Improved |
| replace Qwen3-VL-32B with GPT-5 (ablation) | VLM-driven scenario mining · Iterations | number of iterations: 4.5 -> 3.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- OmniRe — OmniRe: Omni Urban Scene Reconstruction (ICLR 2025) · better on 8 of 8 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SceneCrafter: Controllable Multi-View Driving Scene Editing (CVPR 2025 · similarity 0.88)
- X-Scene: Large-Scale Driving Scene Generation with High Fidelity and Flexible Controllability (NeurIPS 2025 · similarity 0.87)
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · similarity 0.87)
- Editable Scene Simulation for Autonomous Driving via Collaborative LLM-Agents (CVPR 2024 · similarity 0.87)
- Scenario Dreamer: Vectorized Latent Diffusion for Generating Driving Simulation Environments (CVPR 2025 · similarity 0.87)
- One World, Dual Timeline: Decoupled Spatio-Temporal Gaussian Scene Graph for 4D Cooperative Driving Reconstruction (NeurIPS 2026 · 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.
