[GR/MA/SA] Provably Safe Adaptive Reinforcement Learning for Autonomous Driving

Institut

Professur für Cyber Physical Systems (TUM-CIT)

Typ

Semesterarbeit / Masterarbeit /

Inhalt

experimentell /  

Beschreibung

Background

Safe reinforcement learning (RL) has shown strong potential for decision making in complex and uncertain environments. However, for safety-critical systems, achieving high performan-
ce is not sufficient: the learned policy must also remain provably safe under changing environmental conditions. To address this challenge, this thesis focuses on extending a framework that combines learning-based action masking with formally verified fail-safe maneuvers. The method has already been tested in abstract benchmark tasks, to further demonstrate its applicability and effectiveness in more challenging scenarios, we will apply it to autonomous driving scenarios.

 

Description

This thesis aims to extend our current provably safe adaptive RL framework toward autonomous driving scenarios. The student will build on our existing framework and implement longitudinal and lateral action masking networks to prevent driving off-road and guarantee safe-distance with the preceding vehicles. The resulting method will be evaluated under various scenarios, including different road shapes and traffic conditions, and the adaptability will be assessed. The project is closely connected to an ongoing research paper and is intended to provide experimental validation for further publication.

 

Voraussetzungen

Good programming skill of Python, familiar with git, experience with reinforcement learning is a plus, self-motivated working

Möglicher Beginn

sofort

Kontakt

Shuaiyi Li
Raum: 5607.03.035
shuaiyi.litum.de

Ausschreibung