AI Paper Atlas

Research map / Video / Video generation

DriveLaW: Unifying Planning and Video Generation in a Latent Driving World

Tianze Xia, Yongkang Li, Lijun Zhou, Jingfeng Yao, Kaixin Xiong, Haiyang Sun, Bing Wang, Kun Ma, Guang Chen, Hangjun Ye, Wenyu Liu, Xinggang Wang

CVPR 2026 · Video › Video generation · approach: World-model video generation

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 changedWhereMeasuredEffect
condition planner on video-generator latents instead of BEV features (ablation)DriveLaW-Act · NAVSIM NavtestPDMS: 84.1 -> 89.1Improved
condition planner on video latents instead of VLM hidden states (ablation)DriveLaW-Act · NAVSIM NavtestPDMS: 86.5 -> 89.1Improved
add driving-domain video pretraining (7.6M samples) (ablation)DriveLaW video generator · NAVSIM NavtestPDMS: 85.9 -> 89.1Improved
scale video pretraining from 76k to 3.8M samples (ablation)DriveLaW video generator · NAVSIM NavtestPDMS: 87.0 -> 87.8Improved
use latents from later denoising step (t=5) rather than t=1 (ablation)DriveLaW-Act · NAVSIM NavtestPDMS: 89.1 -> 86.9Hurt
use latents from denoising step t=10 (ablation)DriveLaW-Act · NAVSIM NavtestPDMS: 89.1 -> 23.2Hurt
add first-stage training on long low-res clips (ablation)DriveLaW-Video · nuScenes validationFVD: 109.3 -> 81.3Improved
add second-stage high-res training (ablation)DriveLaW-Video · nuScenes validationFVD: 93.2 -> 81.3Improved
add noise reinjection (ablation)DriveLaW-Video · nuScenes validationFVD: 102.1 -> 81.3Improved
add noise reinjection (ablation)DriveLaW-Video · nuScenes validationFID: 6.1 -> 4.6Improved
chain video generation into planner instead of parallel designDriveLaW vs Epona · NAVSIM NavtestPDMS: 86.2 -> 89.1Improved
longer horizon generation (100 frames) versus EponaDriveLaW-Video · NuPlan FVD100FVD: 277.3 -> 296.1Hurt

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.

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

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy