This research introduces a novel framework that uses Large Language Models to automatically generate diverse traffic scenarios for autonomous vehicle testing in the CARLA simulator.

• Creates realistic traffic environments from natural language descriptions
• Includes multi-stage pipeline: prompt analysis, road retrieval, and agent planning
• Enhances testing capabilities by producing diverse, customizable scenarios
• Demonstrates potential for reducing collision rates in autonomous systems

This engineering breakthrough enables more comprehensive testing of autonomous vehicles across varied traffic conditions, helping to identify edge cases and improve safety protocols without manual scenario creation.

Traffic Scene Generation from Natural Language Description for Autonomous Vehicles with Large Language Model