Film mode plays in landscape on mobile. Turn your phone sideways, or tap below to enter landscape fullscreen.
Long-horizon driving world model
HorizonDrive
Self-Corrective Autoregressive World Model for Long-horizon Driving Simulation
Abstract
HorizonDrive keeps autoregressive driving simulation reliable beyond short clips.
Method Overview
We first train a conditional driving world model, then improve its autoregressive stability through scheduled rollout recovery, and finally distill long-horizon teacher rollouts into a few-step, short-chunk student via teacher-rollout DMD.
Results
20-Second AR Results on Nuscenes
2x playbackEach model rollout generates 1 s of future video.
Results
30-Second AR Results on Self-Collected Dataset
2x playbackEach model rollout generates 1 s of future video.
Results
Minute-Level AR Video Generation
3x playbackEach model rollout generates 1 s of future video.
Simulation
Closed-Loop Driving Simulation
Each model rollout generates 1 s of future video.
HorizonDrive
Thank you for your interest in our work.
See more details at