Research map / Video / Video generation
DriveVA: Video Action Models are Zero-Shot Drivers
Mengmeng Liu, Diankun Zhang, Jiuming Liu, Jianfeng Cui, Hongwei Xie, Guang Chen, Hangjun Ye, Michael Yang, Francesco Nex, Hao Cheng
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 video loss (video supervision) (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 71.4 -> 90.9 | Improved |
| add video continuation (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 84.6 -> 90.9 | Improved |
| add CARLA mix training (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 90.5 -> 90.9 | Improved |
| predict video plus action instead of action only (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 47.0 -> 90.9 | Improved |
| use bidirectional video-action attention instead of causal mask (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 90.1 -> 90.9 | Improved |
| increase sampling steps from 1 to 2 (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 13.2 -> 90.9 | Improved |
| increase sampling steps from 2 to 3 (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 90.9 -> 90.9 | No clear effect |
| use 8 future frames instead of 4 (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 82.1 -> 90.9 | Improved |
| use 12 future frames instead of 8 (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 90.9 -> 86.7 | Hurt |
| full fine-tune instead of LoRA on 5B backbone (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 74.9 -> 90.9 | Improved |
| pretrained video prior with full fine-tune instead of training from scratch (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 62.9 -> 90.9 | Improved |
| scale LoRA backbone from 5B to 14B (ablation) | DriveVA · NAVSIM v1 navtest ablation | PDMS: 74.9 -> 80.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Epona — Epona: Autoregressive Diffusion World Model for Autonomous Driving (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- DriveDreamer-Policy: A Geometry-Grounded World-Action Model for Unified Generation and Planning (NeurIPS 2026 · similarity 0.90)
- SAMPO: Scale-wise Autoregression with Motion Prompt for Generative World Models (NeurIPS 2025 · similarity 0.89)
- From Forecasting to Planning: Policy World Model for Collaborative State-Action Prediction (NeurIPS 2025 · similarity 0.89)
- Driving into the Future: Multiview Visual Forecasting and Planning with World Model for Autonomous Driving (CVPR 2024 · similarity 0.89)
- Future-Aware End-to-End Driving: Bidirectional Modeling of Trajectory Planning and Scene Evolution (NeurIPS 2025 · similarity 0.89)
- FutureSightDrive: Thinking Visually with Spatio-Temporal CoT for Autonomous Driving (NeurIPS 2025 · similarity 0.88)
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.
