DriveWorld: 4D Pre-trained Scene Understanding via World Models for Autonomous Driving

TL;DR

DriveWorld introduces a novel 4D scene understanding framework for autonomous driving that leverages world models and spatio-temporal pre-training on multi-camera videos, significantly improving various driving perception tasks.

Contribution

The paper proposes a world model-based 4D representation learning framework, DriveWorld, with a Memory State-Space Model and Task Prompt for autonomous driving pre-training.

Findings

7.5% increase in 3D object detection mAP

3.0% increase in online mapping IoU

5.0% increase in multi-object tracking AMOTA

Abstract

Vision-centric autonomous driving has recently raised wide attention due to its lower cost. Pre-training is essential for extracting a universal representation. However, current vision-centric pre-training typically relies on either 2D or 3D pre-text tasks, overlooking the temporal characteristics of autonomous driving as a 4D scene understanding task. In this paper, we address this challenge by introducing a world model-based autonomous driving 4D representation learning framework, dubbed \emph{DriveWorld}, which is capable of pre-training from multi-camera driving videos in a spatio-temporal fashion. Specifically, we propose a Memory State-Space Model for spatio-temporal modelling, which consists of a Dynamic Memory Bank module for learning temporal-aware latent dynamics to predict future changes and a Static Scene Propagation module for learning spatial-aware latent statics to offer comprehensive scene contexts. We additionally introduce a Task Prompt to decouple task-aware features for various downstream tasks. The experiments demonstrate that DriveWorld delivers promising results on various autonomous driving tasks. When pre-trained with the OpenScene dataset, DriveWorld achieves a 7.5% increase in mAP for 3D object detection, a 3.0% increase in IoU for online mapping, a 5.0% increase in AMOTA for multi-object tracking, a 0.1m decrease in minADE for motion forecasting, a 3.0% increase in IoU for occupancy prediction, and a 0.34m reduction in average L2 error for planning.