"DriveScene: A Comprehensive Video Dataset for Scenario Understanding, Generation, and Explanation in Autonomous Driving Research"
Abstract
"Ensuring autonomous vehicles are trustworthy and transparent is crucial for safety and public trust, especially in high-stakes emergent scenarios. However, existing driving datasets rarely capture these safety-critical situations or provide the detailed annotations needed for explainable decision-making. To address this, we introduce DriveScene, a large-scale driving video dataset covering a diverse range of scenarios from routine driving to safety-critical emergencies and featuring a comprehensive annotation schema that labels each scene\u2019s key objects, their actions, causal factors, and surrounding context. These rich annotations support several key tasks: identifying important objects for interaction-aware modeling, predicting the temporal unfolding of emergent scenarios, generating human-readable explanations of driving behavior, and modeling human perception of scenario risk for human-centered scenario analysis. "
Fields of Study & Subjects
Autonomous Vehicle Technology and Safety
Human-Automation Interaction and Safety
Traffic and Road Safety
Publication details
- Journal / source
- IEEE DataPort
- Publication date
- 31 January 2026
- Document type
- Dataset