Autonomous Driving Planning on WOD-E2E (test)
7.909RFSPoutine-Base
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Poutine-BaseParadigm=AR, Zero-shot=false, TPS=51.2†, Tok/Step=12026.05 | 7.909 | 2.94 | 1.27 | |
| MindVLA-U1 + RLBackbone=Qwen3-VL-2B [34], AD-PT=×2026.05 | 7.87 | 2.66 | 1.09 | |
| Fast-dDrive (+ Inference scaling (N=4))Paradigm=Block Diff., Zero-shot=false, Inference scaling=N=4, TPS=114.7, Tok/Step=2.762026.05 | 7.827 | 2.821 | 1.24 | |
| Fast-dDrive (Scaffold Spec)Paradigm=Block Diff., Zero-shot=false, TPS=210.4, Tok/Step=4.902026.05 | 7.823 | 2.907 | 1.254 | |
| MindVLA-U1Backbone=Qwen3-VL-2B, AD-PT=×2026.05 | 7.77 | 2.67 | 1.16 | |
| HMVLMBackbone=Qwen2.5-VL-3B [33], AD-PT=?2026.05 | 7.74 | 3.07 | 1.33 | |
| DiffusionLTFBackbone=DiffusionDrive [9], AD-PT=?2026.05 | 7.72 | 2.98 | 1.36 | |
| UniPlan (DiffusionDrive)Backbone=DiffusionDrive [9], AD-PT=?2026.05 | 7.69 | 2.99 | 1.31 | |
| dVLM-ADBackbone=LLaDA-V, # Trajectories=29.3k, Zero-shot evaluation=false2025.12 | 7.633 | 3.022 | 1.285 | |
| dVLM-ADParadigm=Diffusion, Zero-shot=false, TPS=35.2, Tok/Step=2.822026.05 | 7.633 | 3.022 | 1.285 | |
| dVLM-ADBackbone=LLaDA-V [60] + SigLIP2 [61], AD-PT=✓2026.05 | 7.63 | 3.02 | 1.29 | |
| AutoVLABackbone=Qwen2.5-VL-3B [33], AD-PT=✓2026.05 | 7.56 | 2.96 | 1.35 | |
| AutoVLABackbone=Qwen2.5-VL, # Trajectories=52.8k, Zero-shot evaluation=false2025.12 | 7.557 | 2.558 | 1.351 | |
| AutoVLAParadigm=AR, Zero-shot=false, TPS=51.2†, Tok/Step=12026.05 | 7.557 | 2.958 | 1.351 | |
| Swin-TrajectoryBackbone=Swin Transformer [57], AD-PT=?2026.05 | 7.54 | 2.81 | 1.21 | |
| NaiveEMMABackbone=Gemini [59], AD-PT=×2026.05 | 7.53 | 3.02 | 1.32 | |
| NaiveEMMABackbone=Gemini, # Trajectories=-, Zero-shot evaluation=false2025.12 | 7.528 | 3.018 | 1.32 | |
| NaiveEMMAParadigm=AR, Zero-shot=false, Tok/Step=12026.05 | 7.528 | 3.018 | 1.32 | |
| ViT-Adapter-GRUBackbone=n/r, AD-PT=?2026.05 | 7.5 | 2.7 | 1.2 | |
| Open-LLaMABackbone=LLaMA-Vision [58], AD-PT=?2026.05 | 7.43 | 3.22 | 1.31 | |
| Open-LLaMABackbone=LLaMA-Vision, # Trajectories=-, Zero-shot evaluation=false2025.12 | 7.429 | 3.217 | 1.314 | |
| LightEMMABackbone=-, # Trajectories=zero-shot, Zero-shot evaluation=true2025.12 | 6.517 | 3.74 | 1.705 | |
| LightEMMAParadigm=AR, Zero-shot=true, Tok/Step=12026.05 | 6.517 | 3.74 | 1.705 | |
| OpenEMMABackbone=-, # Trajectories=zero-shot, Zero-shot evaluation=true2025.12 | 5.158 | 12.476 | 6.684 | |
| OpenEMMAParadigm=AR, Zero-shot=true, Tok/Step=12026.05 | 5.158 | 12.476 | 6.684 |