DrivingDojo Dataset: Advancing Interactive and Knowledge-Enriched Driving World Model

TL;DR

DrivingDojo is a new dataset designed to enhance interactive driving world models by providing diverse, complex driving scenarios and an action instruction following benchmark, enabling better future prediction capabilities.

Contribution

The paper introduces the DrivingDojo dataset, the first tailored for training interactive and knowledge-enriched driving world models, along with a new benchmark for action instruction following.

Findings

DrivingDojo enables improved future prediction accuracy.

The dataset covers diverse driving maneuvers and multi-agent interactions.

Proposed models outperform baselines on the AIF benchmark.

Abstract

Driving world models have gained increasing attention due to their ability to model complex physical dynamics. However, their superb modeling capability is yet to be fully unleashed due to the limited video diversity in current driving datasets. We introduce DrivingDojo, the first dataset tailor-made for training interactive world models with complex driving dynamics. Our dataset features video clips with a complete set of driving maneuvers, diverse multi-agent interplay, and rich open-world driving knowledge, laying a stepping stone for future world model development. We further define an action instruction following (AIF) benchmark for world models and demonstrate the superiority of the proposed dataset for generating action-controlled future predictions.