Research map / 3D vision / 3D generation
UniSim: A Neural Closed-Loop Sensor Simulator
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 actor model (generalized hypernetwork actor representation) (ablation) | NFF base model · PandaSet lane shift and interpolation | FID @2m and PSNR: FID 153.7 -> 84.1; PSNR 24.93 -> 25.80 | Helped |
| add CNN RGB decoder (ablation) | NFF + actor model · PandaSet lane shift | FID @2m: 84.1 -> 78.8 (2m); 111.8 -> 103.3 (3m) | Helped |
| add VGG perceptual and adversarial losses (ablation) | NFF + actor + CNN · PandaSet lane shift and interpolation | FID @2m / LPIPS / PSNR: FID 78.8 -> 74.7; LPIPS 0.341 -> 0.288; PSNR 25.99 -> 25.63 | Mixed |
| UniSim vs. baselines on camera realism | camera simulation · PandaSet interpolation and lane shift (10 scenes) | PSNR, SSIM, LPIPS, FID: PSNR 25.63 vs 24.03 best baseline; FID@2m 74.7 vs 112.6 best baseline | Helped |
| UniSim LiDAR simulation vs LiDARsim | LiDAR simulation · PandaSet log replay | hit rate, median error, intensity RMSE: Hit rate 92.2% -> 99.6%; intensity RMSE 0.091 -> 0.065 | Helped |
| UniSim-rendered data for perception domain gap (Real2Sim/Sim2Real) | BEVFormer detector · PandaSet replay and lane shift | detection mAP: Lane shift Real2Sim 37.0 vs FVS 30.3; Sim2Real 37.1 vs FVS 32.2 | Helped |
| augment real training data with UniSim simulated lane-shift and replay variations | BEVFormer detector · PandaSet detection | mAP: Real 40.9 -> Real+Sim 42.9 (FVS 41.1, Instant-NGP 40.1) | Helped |
| Instant-NGP simulated data used alone for training | BEVFormer detector · PandaSet detection | mAP: Sim 32.4 vs Real 40.9 | Hurt |
| UniSim simulated data only (no real) for training | BEVFormer detector · PandaSet detection | mAP: Sim (UniSim) 41.4 vs Real 40.9 | Helped |
| UniSim vs FVS for open-loop autonomy metrics | BEVFormer + forecasting + planning · PandaSet open-loop | detection agreement, ADE, plan consistency: Det agreement 0.82 vs 0.80; ADE 1.68 vs 2.35; plan consistency 6.09 vs 6.15 | Mixed |
Closest papers by meaning
- NAVSIM: Data-Driven Non-Reactive Autonomous Vehicle Simulation and Benchmarking (NeurIPS 2024 · similarity 0.87)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- DualAD: Disentangling the Dynamic and Static World for End-to-End Driving (CVPR 2024 · similarity 0.86)
- The Road Ahead in Autonomous Driving: The KITScenes Multimodal Dataset (NeurIPS 2026 · similarity 0.86)
- DriveHierarchy: A Benchmark for Diagnosing VLM Driving Capabilities from Open-Loop Understanding to Closed-Loop Execution (NeurIPS 2026 · similarity 0.86)
- Enhancing End-to-End Autonomous Driving with Latent World Model (ICLR 2025 · 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.