Bachelor/Semester/Master thesis: Generation of safety-critical scenarios based on LLMs

14.11.2024, Diplomarbeiten, Bachelor- und Masterarbeiten

Objective The primary objective of this project is to explore how LLMs can be used to generate safety-critical scenarios within the CommonRoad framework.

Background

The validation and development of autonomous driving heavily rely on scenario-based testing with simulation in the loop and hardware in the loop. Tools like CommonRoad, CARLA, and SUMO have become essential in simulating real-world traffic scenarios. While CARLA is commonly used for generating scenarios to train Reinforcement Learning (RL) agents, and SUMO excels in creating urban traffic plans, CommonRoad's scenarios are meticulously designed to represent a wide range of driving situations, including urban, rural, and highway environments. These scenarios incorporate static and dynamic obstacles, road networks, and goal regions, providing a realistic and challenging testing ground for motion planning algorithms.

Despite its robustness, generating new, safety-critical scenarios within CommonRoad remains a complex task. Traditional methods often involve manual design or graphical user interfaces, which can be time-consuming and may not cover all potential edge cases. Leveraging Large Language Models (LLMs) to generate such scenarios could enhance the diversity and criticality of test cases, providing a more comprehensive evaluation of motion planning algorithms.

Objective

The primary objective of this project is to explore how LLMs can be used to generate safety-critical scenarios within the CommonRoad framework. Specifically, the project will:

· Develop a framework to generate scenarios directly with CommonRoad’s domain-specific language (DSL).
· Develop an LLM-based scenario generator using prompt engineering techniques.
· Ensure that the generated scenarios meet predefined safety-critical metrics such as collision probability, near-miss events, and hazard-prone conditions.

We Offer:

· A dynamic and future-oriented research environment.
· Hands-on experience with a state-of-the-art software stack for autonomous driving.
· Opportunity to publish a scientific paper (based on merit).
· Thesis work can be conducted in English or German.

Requirements (What You Should Bring):

· Initiative and a creative, problem-solving mindset
· Excellent English or German proficiency
· Advanced knowledge of Python or C++
· Knowledge of Machine Learning, specifically Large Language Models, is an advantage
· Familiarity with traffic simulation tools like CommonRoad is an advantage
· Experience with software development tools such as Git and Ubuntu is desirable

Work can begin immediately. If you are interested, simply send an email with a brief cover letter explaining what fascinates you about this topic, a current academic transcript, and a resume to yuan_avs.gao@tum.de.

Kontakt: yuan_avs.gao@tum.de