AutoDrive-QA creates a structured benchmark for evaluating autonomous driving systems by automatically transforming open-ended questions into multiple-choice format.

• Addresses the challenge of unreliable evaluations in autonomous driving question answering
• Converts datasets like DriveLM, NuScenes-QA, and LingoQA into measurable MCQ formats
• Systematically assesses perception, prediction, and planning capabilities
• Eliminates dependency on complex metrics or subjective human judgment

This engineering innovation significantly improves the standardization and reliability of autonomous vehicle evaluation frameworks, critical for safety validation and consistent performance assessment.

AutoDrive-QA: Automated Generation of Multiple-Choice Questions for Autonomous Driving Datasets Using Large Vision-Language Models