STAGE: A Stream-Centric Generative World Model for Long-Horizon Driving-Scene Simulation

TL;DR

STAGE is a novel auto-regressive framework for long-horizon driving video generation that improves temporal consistency and reduces error accumulation through hierarchical feature transfer and multi-stage training, enabling the creation of extended high-quality driving videos.

Contribution

The paper introduces STAGE, a new hierarchical, auto-regressive model with multi-phase training that significantly enhances long-horizon driving video synthesis and temporal consistency.

Findings

Outperforms existing methods on Nuscenes dataset

Successfully generates 600-frame high-quality driving videos

Reduces error accumulation in long-horizon video synthesis

Abstract

The generation of temporally consistent, high-fidelity driving videos over extended horizons presents a fundamental challenge in autonomous driving world modeling. Existing approaches often suffer from error accumulation and feature misalignment due to inadequate decoupling of spatio-temporal dynamics and limited cross-frame feature propagation mechanisms. To address these limitations, we present STAGE (Streaming Temporal Attention Generative Engine), a novel auto-regressive framework that pioneers hierarchical feature coordination and multi-phase optimization for sustainable video synthesis. To achieve high-quality long-horizon driving video generation, we introduce Hierarchical Temporal Feature Transfer (HTFT) and a novel multi-stage training strategy. HTFT enhances temporal consistency between video frames throughout the video generation process by modeling the temporal and denoising process separately and transferring denoising features between frames. The multi-stage training strategy is to divide the training into three stages, through model decoupling and auto-regressive inference process simulation, thereby accelerating model convergence and reducing error accumulation. Experiments on the Nuscenes dataset show that STAGE has significantly surpassed existing methods in the long-horizon driving video generation task. In addition, we also explored STAGE's ability to generate unlimited-length driving videos. We generated 600 frames of high-quality driving videos on the Nuscenes dataset, which far exceeds the maximum length achievable by existing methods.