Planning on NAVSIM (test)
93.5PDMSDriveSuprim
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| DriveSuprimSensor=multi-camera (C), Input Type=end-to-end2026.06 | 93.5 | 98.6 | 98.6 | 95.5 | 100 | 91.3 | |
| Unified Driving TokensSensor=single-view (1V), Input Type=frozen tokenizer2026.06 | 91.8 | 98.7 | 98.2 | 95.9 | 100 | 87.3 | |
| Hydra-MDP++Sensor=cameras with LiDAR (C+L), Input Type=end-to-end2026.06 | 91 | 98.6 | 98.6 | 95.1 | 100 | 85.7 | |
| GT Future (Oracle)Type=Ground-truth + IDM, Multiple Sensors=false, Ego Status=false, Past Traj.=false, Extra Anno.=false2025.06 | 90.8 | — | — | — | — | — | |
| ReWorldImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 90.4 | 99.1 | 98.2 | 97.7 | 99.8 | 82 | |
| DriveVLA-W0Sensor=single-view (1V), Input Type=frozen tokenizer2026.06 | 90.2 | 98.7 | 99.1 | 95.3 | 99.3 | 83.3 | |
| DriveLaWSensor=single-view (1V), Input Type=frozen tokenizer2026.06 | 89.1 | 99 | 97.1 | 96.7 | 100 | 81.3 | |
| DriveLaWImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 89.1 | 99 | 97.1 | 96.7 | 100 | 81.3 | |
| ResWorldSensor=multi-camera (C), Input Type=end-to-end2026.06 | 89 | 98.9 | 96.5 | 95.6 | 100 | 83.1 | |
| WorldDriveImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 89 | 98.4 | 96.8 | 95.2 | 100 | 83.3 | |
| WoTESensor=cameras with LiDAR (C+L), Input Type=end-to-end2026.06 | 88.3 | 98.5 | 96.8 | 94.9 | 99.9 | 81.9 | |
| WoTEImage input=true, Lidar input=true, World Model integration=true, Flow-matching objective=false2026.06 | 88.3 | 98.5 | 96.8 | 94.9 | 99.9 | 81.9 | |
| DiffusionDrivesource=reported in Liao et al. (2025)2025.09 | 88.1 | 98.2 | 96.2 | 94.7 | 100 | 82.2 | |
| DiffusionDriveInput=3Cam+L2026.02 | 88.1 | 96.8 | 95.4 | 94.7 | 100 | 82 | |
| DiffusionDriveSensor=cameras with LiDAR (C+L), Input Type=end-to-end2026.06 | 88.1 | 98.2 | 96.2 | 94.7 | 100 | 82.2 | |
| PWMSensor=single-view (1V), Input Type=frozen tokenizer2026.06 | 88.1 | 98.6 | 95.9 | 95.4 | 100 | 81.8 | |
| DiffusionDriveImage input=true, Lidar input=true, World Model integration=false, Flow-matching objective=false2026.06 | 88.1 | 98.2 | 96.2 | 94.7 | 100 | 82.2 | |
| PWMImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 88.1 | 98.6 | 95.9 | 95.4 | 100 | 81.8 | |
| DRAEInput=C & L, Img. Backbone=ResNet-34, Anchor=202025.07 | 88 | 98.4 | 96.2 | 94.9 | 100 | 82.5 | |
| DRAEInput=C&L, Img. Backbone=ResNet-34, Anchor=202025.07 | 88 | 98.4 | 96.2 | 94.9 | 100 | 82.5 | |
| BridgeDrive2025.09 | 88 | 98.2 | 96.1 | 94.5 | 100 | 82.3 | |
| DiffusionDrivevariant=reproduced with a different seed2025.09 | 87.6 | 98.2 | 95.9 | 94.3 | 100 | 81.9 | |
| InternVL2.5 + ImagiDrive-SInput=Camera2025.08 | 87.4 | 98.6 | 96.2 | 94.5 | 100 | 80.5 | |
| DriveVLA-W0Image input=true, Lidar input=false, World Model integration=true, Flow-matching objective=true2026.06 | 87.2 | 98.4 | 95.3 | 95.2 | 100 | 80.9 | |
| InternVL2.5 + ImagiDrive-AInput=Camera2025.08 | 86.9 | 98.1 | 96.2 | 94.4 | 100 | 80.1 | |
| ReSim + IDMType=WM + IDM, Multiple Sensors=false, Ego Status=false, Past Traj.=false, Extra Anno.=false2025.06 | 86.6 | — | — | — | — | — | |
| ReSimImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 86.6 | — | — | — | — | — | |
| Hydra-MDP-V8192-W-EP2025.09 | 86.5 | 98.3 | 96 | 94.6 | 100 | 78.7 | |
| ReCogDrive-ILImage input=true, Lidar input=false, World Model integration=false, Flow-matching objective=false2026.06 | 86.5 | 98.1 | 94.7 | 94.2 | 100 | 80.9 | |
| LLava-1.6 + ImagiDrive-SInput=Camera2025.08 | 86.4 | 97.9 | 95.5 | 93.1 | 99.9 | 80.7 | |
| EponaInput=Camera2025.08 | 86.2 | 97.9 | 95.1 | 93.8 | 99.9 | 80.4 | |
| EponaSensor=single-view (1V), Input Type=frozen tokenizer2026.06 | 86.2 | 97.9 | 95.1 | 93.8 | 99.9 | 80.4 | |
| EponaImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 86.2 | 97.9 | 95.1 | 93.8 | 99.9 | 80.4 | |
| LLava-1.6 + ImagiDrive-AInput=Camera2025.08 | 86 | 97.7 | 95.3 | 93 | 99.9 | 80.4 | |
| DRAMAInput=C & L, Img. Backbone=ResNet-34, Anchor=02025.07 | 85.5 | 98 | 93.1 | 94.8 | 100 | 80.1 | |
| DRAMAInput=C&L, Img. Backbone=ResNet-34, Anchor=02025.07 | 85.5 | 98 | 93.1 | 94.8 | 100 | 80.1 | |
| DRAMAInput=Camera & Lidar2025.08 | 85.5 | 98 | 93.1 | 94.8 | 100 | 80.1 | |
| DRAMASensor=cameras with LiDAR (C+L), Input Type=end-to-end2026.06 | 85.5 | 98 | 93.1 | 94.8 | 100 | 80.1 | |
| LFGInput=1Cam*2026.02 | 85.2 | 98.2 | 93.7 | 94.4 | 100 | 79.1 | |
| World4DriveInput=Camera2025.08 | 85.1 | 97.4 | 94.3 | 92.8 | 100 | 79.9 | |
| Hydra-MDPInput=3Cam+L2026.02 | 84.7 | 96.9 | 94 | 94 | 100 | 78.7 | |
| LAWType=WM + E2E Plan, Multiple Sensors=true, Ego Status=false, Past Traj.=false, Extra Anno.=false2025.06 | 84.6 | — | — | — | — | — | |
| PARA-DriveInput=Camera2025.08 | 84.6 | 97.9 | 92.4 | 93 | 99.8 | 79.3 | |
| LAWImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 84.6 | 96.4 | 95.4 | 88.7 | 99.9 | 81.7 | |
| TransfuserType=E2E Plan, Multiple Sensors=true, Ego Status=true, Past Traj.=false, Extra Anno.=true2025.06 | 84 | — | — | — | — | — | |
| PARA-DriveInput=Camera, Img. Backbone=ResNet-34, Anchor=02025.07 | 84 | 97.9 | 92.4 | 93 | 99.8 | 79.3 | |
| TransfuserInput=C & L, Img. Backbone=ResNet-34, Anchor=02025.07 | 84 | 97.7 | 92.8 | 92.8 | 100 | 79.2 | |
| PARA-DriveInput=Cam, Img. Backbone=ResNet-34, Anchor=02025.07 | 84 | 97.9 | 92.4 | 93 | 99.8 | 79.3 | |
| TransfuserInput=C&L, Img. Backbone=ResNet-34, Anchor=02025.07 | 84 | 97.7 | 92.8 | 92.8 | 100 | 79.2 | |
| TransFuserInput=Camera & Lidar2025.08 | 84 | 97.7 | 92.8 | 92.8 | 100 | 79.2 | |
| TransFuserInput=3Cam+L2026.02 | 84 | 97.7 | 92.8 | 92 | 100 | 79.2 | |
| Para-DriveSensor=multi-camera (C), Input Type=end-to-end2026.06 | 84 | 97.9 | 92.4 | 93 | 99.8 | 79.3 | |
| TransFuserSensor=cameras with LiDAR (C+L), Input Type=end-to-end2026.06 | 84 | 97.7 | 92.8 | 92.8 | 100 | 79.2 | |
| TransFuserImage input=true, Lidar input=true, World Model integration=false, Flow-matching objective=false2026.06 | 84 | 97.7 | 92.8 | 92.8 | 100 | 79.2 | |
| PARA-DriveImage input=true, Lidar input=false, World Model integration=false, Flow-matching objective=false2026.06 | 84 | 97.9 | 92.4 | 93 | 99.8 | 79.3 | |
| LTFInput=Camera, Img. Backbone=ResNet-34, Anchor=02025.07 | 83.8 | 97.4 | 92.8 | 92.4 | 100 | 79 | |
| LTFInput=Cam, Img. Backbone=ResNet-34, Anchor=02025.07 | 83.8 | 97.4 | 92.8 | 92.4 | 100 | 79 | |
| UniADType=E2E Plan, Multiple Sensors=true, Ego Status=true, Past Traj.=false, Extra Anno.=true2025.06 | 83.4 | — | — | — | — | — | |
| UniADInput=Camera, Img. Backbone=ResNet-34, Anchor=02025.07 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| UniADInput=Cam, Img. Backbone=ResNet-34, Anchor=02025.07 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| UniADInput=Camera2025.08 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| UniADInput=6Cam2026.02 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| UniADSensor=multi-camera (C), Input Type=end-to-end2026.06 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| UniADImage input=true, Lidar input=false, World Model integration=false, Flow-matching objective=false2026.06 | 83.4 | 97.8 | 91.9 | 92.9 | 100 | 78.8 | |
| DrivingGPTType=WM + E2E Plan, Multiple Sensors=false, Ego Status=false, Past Traj.=true, Extra Anno.=false2025.06 | 82.4 | — | — | — | — | — | |
| DrivingGPTego status=false, history observation duration=2 seconds, prediction horizon=4 seconds2024.12 | 82.4 | 98.9 | 90.7 | 94.9 | 95.6 | 79.7 | |
| DrivingGPTInput=Camera2025.08 | 82.4 | 98.9 | 90.7 | 94.9 | 95.6 | 79.7 | |
| DrivingGPTImage input=true, Lidar input=false, World Model integration=true, Flow-matching objective=false2026.06 | 82.4 | 98.9 | 90.7 | 94.9 | 95.6 | 79.7 | |
| VADv2Input=Camera & Lidar2025.08 | 80.9 | 97.2 | 89.1 | 91.9 | 100 | 76 | |
| VADv2-V81922025.09 | 80.9 | 97.2 | 89.1 | 91.6 | 100 | 76 | |
| VADv2Sensor=multi-camera (C), Input Type=end-to-end2026.06 | 80.9 | 97.2 | 89.1 | 91.6 | 100 | 76 | |
| VADv2-V8192Image input=true, Lidar input=false, World Model integration=false, Flow-matching objective=false2026.06 | 80.9 | 97.2 | 89.1 | 91.6 | 100 | 76 | |
| VO plannerType=E2E Plan, Multiple Sensors=false, Ego Status=false, Past Traj.=false, Extra Anno.=false2025.06 | 78.4 | — | — | — | — | — | |
| ResNet-50 + MLP baselineBackbone=ResNet-50, Trajectory Decoder=MLP, input=front camera image, ego status=false2024.12 | 77.8 | 92.6 | 89.9 | 86.2 | 96.3 | 73.7 | |
| Constant Velocity2024.12 | 24.2 | 66.7 | 63.9 | 45.2 | 100 | 23.6 |