As autonomous vehicles mature, perception alone is not enough. They must understand context, follow nuanced traffic laws, explain their actions, and adapt to dynamic environments—capabilities that demand real-time intelligence beyond sensor data.
This session explores how Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) pipelines are revolutionizing the reasoning layer of autonomous driving systems. We’ll deep dive into the full architecture that enables natural language querying of map data, traffic law compliance, real-time diagnostics, and human-readable decision explanations.
Topics include:
Designing RAG pipelines for vehicle-level data (logs, GPS, diagnostics)
Integrating vector databases (FAISS/Pinecone) with LangChain or LlamaIndex
Prompt engineering for traffic-specific queries
Evaluating LLMs (Claude, LLaMA, Mistral) in low-latency use cases
Challenges: latency, hallucinations, retrieval precision
Deployment strategies using FastAPI + Docker for edge/cloud inference
This talk is deeply technical and hands-on—ideal for AI engineers, autonomous system designers, and technical leads aiming to build the next generation of knowledge-aware AVs that don’t just drive but explain.