This research introduces ADS-Edit, a specialized dataset enabling targeted modifications to Large Multimodal Models (LMMs) for autonomous driving without complete retraining.

• Addresses key challenges in applying LMMs to autonomous driving, including misunderstood traffic rules and complex road conditions
• Utilizes Knowledge Editing techniques to make precise adjustments to model behavior
• Creates a multimodal dataset specifically designed for autonomous vehicle applications
• Improves safety and reliability of self-driving systems through targeted knowledge enhancement

This engineering advancement matters because it offers a practical path to refine autonomous driving systems without resource-intensive retraining, potentially accelerating deployment of safer self-driving technologies.

ADS-Edit: A Multimodal Knowledge Editing Dataset for Autonomous Driving Systems