Bench2Drive is a popular autonomous driving benchmark. Over 40 methods have been tested on Bench2Drive. Papers using the benchmark usually only report a few baselines due to space constraints making it hard to keep an overview over the literature. To solve this, this repository contains a complete list of all methods that have currently been tested on Bench2Drive.
The results are from Bench2Drive version 0.0.3 and self-reported, copied from the respective publications. Methods are sorted by Driving Score (DS). At the bottom of the table, separated by a line, are privileged methods that use ground truth perception as input. This leaderboard compares the final performance of entire methods, including data acquisition methods, not just architectures. Methods vary among many axes: driving dataset used, pre-training strategy, training compute, inference compute, number of parameters, IL/RL, etc. In particular, many (4/5) of the current top-performing methods use the PDM-Lite dataset.
To add a new method please open a pull request. For fixing errors please open a GitHub issue.
Main.tex contains the leaderboard table as latex with the associated citations in CVPR-style.
Last updated: 10. Nov. 2025
| Method | DS ↑ | SR ↑ | Year | Venue |
|---|---|---|---|---|
| AD-MLP | 18.05 | 0.00 | 2023 | - |
| TCP | 40.70 | 15.00 | 2022 | NeurIPS |
| ReAL-AD | 41.17 | 11.36 | 2025 | - |
| VAD | 42.35 | 15.00 | 2023 | ICCV |
| SparseDrive | 44.54 | 16.71 | 2025 | ICRA |
| GenAD | 44.81 | 15.90 | 2024 | ECCV |
| Dual-AEB | 45.23 | 10.00 | 2025 | ICRA |
| FUMP | 45.67 | 16.36 | 2025 | - |
| FocalAD | 45.77 | 17.30 | 2025 | - |
| UniAD | 45.81 | 16.36 | 2023 | CVPR |
| MomAD | 47.91 | 18.11 | 2025 | CVPR |
| CogAD | 48.30 | 24.00 | 2025 | - |
| BridgeAD | 50.06 | 22.73 | 2025 | CVPR |
| E³AD | 50.07 | 20.12 | 2025 | - |
| X-Driver | 51.70 | 18.10 | 2025 | - |
| DiFSD | 52.02 | 21.00 | 2024 | - |
| SpaRC-AD | 55.60 | 30.00 | 2025 | Workshop |
| SeerDrive | 58.32 | 30.17 | 2025 | NeurIPS |
| WoTE | 61.71 | 31.36 | 2025 | ICCV |
| DriveDPO | 62.02 | 30.62 | 2025 | NeurIPS |
| ThinkTwice | 62.44 | 31.23 | 2022 | CVPR |
| DriveTransformer-L. | 63.46 | 35.01 | 2025 | ICLR |
| DriveAdapter | 64.22 | 33.08 | 2023 | ICCV |
| VeteranAD | 64.22 | 33.85 | 2025 | - |
| iPad | 65.02 | 35.91 | 2025 | - |
| StuckSolver | 65.23 | 36.32 | 2025 | - |
| GEMINUS | 65.39 | 37.73 | 2025 | - |
| VDRive | 66.25 | 50.51 | 2025 | - |
| RAP-ResNet | 66.42 | 37.27 | 2025 | - |
| CAPS | 66.76 | 52.87 | 2025 | - |
| SNG | 67.17 | 35.90 | 2025 | - |
| DiffAD | 67.92 | 38.64 | 2025 | - |
| OAIAD | 68.73 | 48.86 | 2025 | Machines |
| DIVER | 68.90 | 36.75 | 2025 | - |
| Raw2Drive | 71.36 | 50.24 | 2025 | NeurIPS |
| VL | 73.29 | 65.44 | 2025 | ICRA |
| Hydra-NeXt | 73.86 | 50.00 | 2025 | - |
| DriveMoE | 74.22 | 48.64 | 2025 | - |
| ETA | 74.33 | 48.33 | 2025 | ICCV |
| DiffusionDrive-temp | 77.68 | 52.72 | 2025 | CVPR |
| ORION | 77.74 | 54.62 | 2025 | ICCV |
| PGS | 78.08 | 48.64 | 2025 | NeurIPS |
| GaussianFusion | 79.10 | 54.40 | 2025 | NeurIPS |
| TF++ | 84.21 | 67.27 | 2024 | - |
| SimLingo | 85.07 | 67.27 | 2025 | CVPR |
| R2SE | 86.28 | 69.54 | 2025 | - |
| HiP-AD | 86.77 | 69.09 | 2025 | ICCV |
| BridgeDrive | 86.87 | 72.27 | 2025 | - |
| --- | --- | --- | --- | --- |
| Think2Drive | 91.85 | 85.41 | 2024 | ECCV |
| PDM-Lite | 97.02 | 92.27 | 2024 | ECCV |