This research introduces Driving-RAG, a novel framework that improves autonomous vehicles' decision-making by efficiently retrieving and learning from previous driving scenarios.

• Creates high-quality scenario embeddings that capture critical driving features
• Enables semantic search across massive driving datasets with 95% accuracy
• Demonstrates effective RAG applications that enhance trajectory planning and decision-making
• Provides a comprehensive benchmark dataset for driving scenario retrieval evaluation

For engineering teams, this technology represents a significant advancement that allows autonomous driving systems to leverage past experiences for safer, more intelligent real-time decisions in complex environments.

Driving-RAG: Driving Scenarios Embedding, Search, and RAG Applications