CurricuVLM integrates Vision-Language Models with personalized curriculum learning to enhance autonomous vehicle safety in rare but dangerous scenarios.

• Adaptive training approach that identifies and focuses on an AV's specific performance weaknesses
• Vision-Language Model integration enables more natural scenario understanding and generation
• Personalized safety curriculum that evolves with the vehicle's learning progress
• Demonstrated improvements in handling safety-critical edge cases

This research addresses a critical gap in autonomous driving security by systematically training vehicles to handle potential catastrophic scenarios before deployment on public roads.

CurricuVLM: Towards Safe Autonomous Driving via Personalized Safety-Critical Curriculum Learning with Vision-Language Models