Closed-loop Planning on Bench2Drive
91.01Driving ScoreLinkVLA
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| LinkVLAExpert=PDM-Lite, VLA=true, Generative Planning=false2026.07 | 91.01 | 74.55 | 255.84 | 34.62 | — | — | — | — | — | — | |
| AnchorVLAExpert=PDM-Lite, VLA=true, Generative Planning=true2026.07 | 89.92 | 77.28 | 251.14 | 28.94 | — | — | — | — | — | — | |
| SparseDriveV22026.03 | 89.15 | 70 | 199.84 | 18.32 | — | — | — | — | — | — | |
| SparseDriveV2Source=SparseDriveV2 T2+T42026.04 | 89.15 | 70 | — | — | — | — | — | — | — | — | |
| AlignDriveExpert=Think2Drive (Li et al., 2024a)2026.01 | 89.07 | 73.18 | 212.07 | 16.86 | — | — | — | — | — | — | |
| BridgeDriveExpert=PDM-Lite, VLA=false, Generative Planning=true2026.07 | 87.99 | 74.99 | 236.49 | 20.98 | — | — | — | — | — | — | |
| DiffRefinerSource=Abstract (AAAI’26)2026.04 | 87.1 | 71.4 | — | — | — | — | — | — | — | — | |
| HiP-ADParadigm=Traditional Modular Paradigm2025.12 | 86.77 | 69.09 | — | — | — | — | — | — | — | — | |
| HiP-ADParadigm=Traditional Modular2025.12 | 86.77 | 69.09 | — | — | — | — | — | — | — | — | |
| HiP-ADExpert=Think2Drive (Li et al., 2024a)2026.01 | 86.77 | 69.09 | 203.12 | 19.36 | — | — | — | — | — | — | |
| HiP-AD2026.03 | 86.77 | 69.09 | 203.12 | 19.36 | — | — | — | — | — | — | |
| HiP-ADSource=SparseDriveV2 T42026.04 | 86.77 | 69.09 | — | — | — | — | — | — | — | — | |
| HiP-ADParams=98M, Category=Traditional2026.06 | 86.77 | 69.09 | — | — | — | — | — | — | — | — | |
| R2SEParadigm=Traditional Modular2025.12 | 86.28 | 67.76 | — | — | — | — | — | — | — | — | |
| R2SETraining Expert=PDM-Lite2026.04 | 86.28 | 69.54 | 243.89 | 23.26 | — | — | — | — | — | — | |
| SimLingoExpert=PDM-Lite (Beißwenger, 2024)2026.01 | 86.02 | 67.27 | 259.23 | 33.67 | — | — | — | — | — | — | |
| SimLingo2026.03 | 86.02 | 67.27 | 259.23 | 33.67 | — | — | — | — | — | — | |
| SimLingoSource=SparseDriveV2 T42026.04 | 86.02 | 67.27 | — | — | — | — | — | — | — | — | |
| SimLingoTraining Expert=PDM-Lite2026.04 | 85.94 | 66.82 | 244.18 | 25.49 | — | — | — | — | — | — | |
| SimLingoParadigm=VLM-based Paradigm2025.12 | 85.07 | 67.27 | — | — | — | — | — | — | — | — | |
| SimLingoParadigm=VLM-based2025.12 | 85.07 | 67.27 | — | — | — | — | — | — | — | — | |
| SimLingoLLM=InternVL2-1B, Params=1.36B, Category=2D MLLM, Additional training data=true2026.06 | 85.07 | 67.27 | — | — | — | — | — | — | — | — | |
| SimLingoExpert=PDM-Lite, VLA=true, Generative Planning=false2026.07 | 85.07 | 67.27 | 259.23 | 33.67 | — | — | — | — | — | — | |
| TF++Paradigm=Traditional Modular2025.12 | 84.21 | 64.39 | — | — | — | — | — | — | — | — | |
| TransFuser++Expert=PDM-Lite, VLA=false, Generative Planning=false2026.07 | 84.21 | 67.27 | — | — | — | — | — | — | — | — | |
| DriveSuprimSource=DriveSuprim T2+T42026.04 | 83.02 | 60 | — | — | — | — | — | — | — | — | |
| OmniSpaceLLM=Qwen3-VL 4B, Params=4.44B, Category=2D MLLM + train-time 3D model (Ours)2026.06 | 81.4 | 65.52 | — | — | — | — | — | — | — | — | |
| DiffusionDriveExpert=PDM-Lite, VLA=false, Generative Planning=true2026.07 | 80.79 | 58.18 | 248.18 | 24.56 | — | — | — | — | — | — | |
| OmniSpaceLLM=Qwen2.5-VL 7B, Params=7.79B, Category=2D MLLM + train-time 3D model (Ours)2026.06 | 79.65 | 60.15 | — | — | — | — | — | — | — | — | |
| GaussianFusionParadigm=Traditional Modular2025.12 | 79.1 | 54.4 | — | — | — | — | — | — | — | — | |
| AutoVLATraining Expert=PDM-Lite2026.04 | 78.84 | 57.73 | 146.93 | 39.33 | — | — | — | — | — | — | |
| AutoVLAExpert=PDM-Lite, VLA=true, Generative Planning=false2026.07 | 78.84 | 57.73 | 146.93 | 39.33 | — | — | — | — | — | — | |
| UniDriveVLATraining Expert=Think2Drive2026.04 | 78.37 | 51.82 | 198.86 | 11.78 | — | — | — | — | — | — | |
| PGSParadigm=Traditional Modular2025.12 | 78.08 | 48.64 | — | — | — | — | — | — | — | — | |
| SpaceDrive+Paradigm=VLM-based Paradigm2025.12 | 78.02 | 55.11 | — | — | — | — | — | — | — | — | |
| SpaceDrive+Paradigm=VLM-based2025.12 | 78.02 | 55.11 | — | — | — | — | — | — | — | — | |
| SpaceDriveLLM=Qwen2.5-VL 7B, Params=8.7B, Category=2D MLLM + test-time 3D model2026.06 | 78.02 | 55.11 | — | — | — | — | — | — | — | — | |
| ORIONParadigm=VLM-based Paradigm2025.12 | 77.74 | 54.62 | — | — | — | — | — | — | — | — | |
| ORIONParadigm=VLM-based2025.12 | 77.74 | 54.62 | — | — | — | — | — | — | — | — | |
| OrionTraining Expert=Think2Drive2026.04 | 77.74 | 54.62 | 151.48 | 17.38 | — | — | — | — | — | — | |
| ORIONLLM=Vicuna-7B-v1.5, Params=7.55B, Category=2D MLLM + test-time 3D model2026.06 | 77.74 | 54.62 | — | — | — | — | — | — | — | — | |
| Orion-VAEExpert=Think2Drive, VLA=true, Generative Planning=false2026.07 | 77.74 | 54.62 | 151.48 | 17.38 | — | — | — | — | — | — | |
| DiffusionDriveParadigm=Traditional Modular2025.12 | 77.68 | 52.72 | — | — | — | — | — | — | — | — | |
| VADv2Source=VADv2 T2+T52026.04 | 76.15 | 50.46 | — | — | — | — | — | — | — | — | |
| GuideFlowSensor=2 Cams, Expert feature distillation=false2025.11 | 75.21 | 51.36 | — | — | — | — | — | — | — | — | |
| VLR-DriveParadigm=VLM-based Paradigm2025.12 | 75.01 | 50 | — | — | — | — | — | — | — | — | |
| VLR-DriveParadigm=VLM-based2025.12 | 75.01 | 50 | — | — | — | — | — | — | — | — | |
| VLR-DriveLLM=LLaVA-NeXT Video 7B, Category=2D MLLM2026.06 | 75.01 | 50 | — | — | — | — | — | — | — | — | |
| ETAParadigm=VLM-based Paradigm2025.12 | 74.33 | 48.33 | — | — | — | — | — | — | — | — | |
| ETAParadigm=VLM-based2025.12 | 74.33 | 48.33 | — | — | — | — | — | — | — | — | |
| ETALLM=LLaVA 1.6 7B Vicuna, Category=2D MLLM2026.06 | 74.33 | 48.33 | — | — | — | — | — | — | — | — | |
| DriveMoEParadigm=VLM-based Paradigm2025.12 | 74.22 | 48.64 | — | — | — | — | — | — | — | — | |
| DriveMoEParadigm=VLM-based2025.12 | 74.22 | 48.64 | — | — | — | — | — | — | — | — | |
| DriveMOETraining Expert=Think2Drive2026.04 | 74.22 | 48.64 | 175.96 | 15.31 | — | — | — | — | — | — | |
| DriveMoELLM=Paligemma-3B, Category=2D MLLM2026.06 | 74.22 | 48.64 | — | — | — | — | — | — | — | — | |
| Hydra-NeXtParadigm=Traditional Modular2025.12 | 73.86 | 53.22 | — | — | — | — | — | — | — | — | |
| Hydra-NeXtExpert=Think2Drive (Li et al., 2024a)2026.01 | 73.86 | 50 | 197.76 | 20.68 | — | — | — | — | — | — | |
| Hydra-NextSensor=2 Cams, Expert feature distillation=false2025.11 | 73.86 | 50 | — | — | — | — | — | — | — | — | |
| Hydra-NeXt2026.03 | 73.86 | 50 | 197.76 | 20.68 | — | — | — | — | — | — | |
| Hydra-NeXtSource=Hydra-NeXt T2+T42026.04 | 73.86 | 50 | — | — | — | — | — | — | — | — | |
| Orion-DiffusionExpert=Think2Drive, VLA=true, Generative Planning=true2026.07 | 71.97 | 46.54 | — | — | — | — | — | — | — | — | |
| Raw2DriveParadigm=Traditional Modular2025.12 | 71.36 | 50.24 | — | — | — | — | — | — | — | — | |
| ReCogDriveTraining Expert=Think2Drive2026.04 | 71.36 | 45.45 | 138.18 | 17.45 | — | — | — | — | — | — | |
| StuckSolverParadigm=VLM-based Paradigm2025.12 | 70.89 | 50.01 | — | — | — | — | — | — | — | — | |
| StuckSolverParadigm=VLM-based2025.12 | 70.89 | 50.01 | — | — | — | — | — | — | — | — | |
| StuckSolverLLM=GPT-4o, Category=2D MLLM2026.06 | 70.89 | 50.01 | — | — | — | — | — | — | — | — | |
| SafeDriveSource=SafeDrive T1+T32026.04 | 66.77 | 42.4 | — | — | — | — | — | — | — | — | |
| RAPTraining Expert=Think2Drive2026.04 | 66.42 | 37.27 | 165.47 | 23.63 | — | — | — | — | — | — | |
| VDRiveParadigm=VLM-based Paradigm2025.12 | 66.25 | 50.51 | — | — | — | — | — | — | — | — | |
| VDRiveParadigm=VLM-based2025.12 | 66.25 | 50.51 | — | — | — | — | — | — | — | — | |
| VDRiveLLM=InternVL3-8B, Category=2D MLLM2026.06 | 66.25 | 50.51 | — | — | — | — | — | — | — | — | |
| GEMINUSParadigm=VLM-based Paradigm2025.12 | 65.39 | 37.73 | — | — | — | — | — | — | — | — | |
| GEMINUSParadigm=VLM-based2025.12 | 65.39 | 37.73 | — | — | — | — | — | — | — | — | |
| GEMINUSCategory=2D MLLM2026.06 | 65.39 | 37.73 | — | — | — | — | — | — | — | — | |
| DriveAdapterExpert feature distillation=true2025.05 | 64.22 | 33.08 | 70.22 | 16.01 | 14.55 | 22.61 | 54.04 | 50 | 50.45 | 38.33 | |
| DriveAdapterParadigm=Traditional Modular Paradigm2025.12 | 64.22 | 42.08 | — | — | — | — | — | — | — | — | |
| DriveAdapterParadigm=Traditional Modular2025.12 | 64.22 | 42.08 | — | — | — | — | — | — | — | — | |
| DriveAdapterExpert=Think2Drive (Li et al., 2024a), Expert feature distillation=true2026.01 | 64.22 | 33.08 | 70.22 | 16.01 | — | — | — | — | — | — | |
| DriveAdapterSensor=6 Cams, Expert feature distillation=true2025.11 | 64.22 | 33.08 | — | — | — | — | — | — | — | — | |
| DriveAdapterexpert feature distillation=true2026.03 | 64.22 | 33.08 | 70.22 | 16.01 | — | — | — | — | — | — | |
| DriveAdapterTraining Expert=Think2Drive2026.04 | 64.22 | 33.08 | 70.22 | 16.01 | — | — | — | — | — | — | |
| DriveAdapterParams=161M, Category=Traditional2026.06 | 64.22 | 42.08 | — | — | — | — | — | — | — | — | |
| DriveAdapterExpert=Think2Drive, VLA=false, Generative Planning=false2026.07 | 64.22 | 33.08 | 70.22 | 16.01 | — | — | — | — | — | — | |
| ReasonPlan2025.05 | 64.01 | 34.55 | 180.64 | 25.63 | 37.5 | 26.67 | 33.3 | 40 | 45.79 | 36.66 | |
| DriveTransformer-LParadigm=Traditional Modular Paradigm2025.12 | 63.46 | 38.6 | — | — | — | — | — | — | — | — | |
| DriveTransformer-LParadigm=Traditional Modular2025.12 | 63.46 | 38.6 | — | — | — | — | — | — | — | — | |
| DriveTransformerExpert=Think2Drive (Li et al., 2024a)2026.01 | 63.46 | 35.01 | 100.64 | 20.78 | — | — | — | — | — | — | |
| DriveTransformer2026.03 | 63.46 | 35.01 | 100.64 | 20.78 | — | — | — | — | — | — | |
| DriveTransformerTraining Expert=Think2Drive2026.04 | 63.46 | 35.01 | 100.64 | 20.78 | — | — | — | — | — | — | |
| DriveTransformerSource=SparseDriveV2 T42026.04 | 63.46 | 35.01 | — | — | — | — | — | — | — | — | |
| DriveTransformer-LParams=520M, Category=Traditional2026.06 | 63.46 | 38.6 | — | — | — | — | — | — | — | — | |
| DriveTransformerExpert=Think2Drive, VLA=false, Generative Planning=false2026.07 | 63.46 | 35.01 | 100.64 | 20.78 | — | — | — | — | — | — | |
| ThinkTwiceExpert feature distillation=true2025.05 | 62.44 | 31.23 | 69.33 | 16.22 | 13.72 | 22.93 | 52.99 | 50 | 47.78 | 37.48 | |
| ThinkTwiceParadigm=Traditional Modular Paradigm2025.12 | 62.44 | 37.17 | — | — | — | — | — | — | — | — | |
| ThinkTwiceParadigm=Traditional Modular2025.12 | 62.44 | 37.17 | — | — | — | — | — | — | — | — | |
| ThinkTwiceExpert=Think2Drive (Li et al., 2024a), Expert feature distillation=true2026.01 | 62.44 | 31.23 | 69.33 | 16.22 | — | — | — | — | — | — | |
| ThinkTwiceSensor=6 Cams, Expert feature distillation=true2025.11 | 62.44 | 31.23 | — | — | — | — | — | — | — | — | |
| ThinkTwiceexpert feature distillation=true2026.03 | 62.44 | 31.23 | 69.33 | 16.22 | — | — | — | — | — | — | |
| ThinkTwiceTraining Expert=Think2Drive2026.04 | 62.44 | 31.23 | 69.33 | 16.22 | — | — | — | — | — | — | |
| ThinkTwiceParams=129M, Category=Traditional2026.06 | 62.44 | 37.17 | — | — | — | — | — | — | — | — |