MS Thesis in collaboration with DENSO Automotive

Master Thesis: Integrating World Models with AI-based Autonomous Driving

01.04.2026, Diplomarbeiten, Bachelor- und Masterarbeiten

Current AI models show strong preception and prediciton capabilities for autonomous driving, but often lack precise metric and geometric grounding, such as distances, occupancy, and physical constraints. This can lead to geometrically inconsistent or physically implausible action predictions. Recent work on world models are addressing this limitations and are focus of this thesis.

This master’s thesis focuses on Complementing AI models with geometry‑aware World Models that explicitly encode metric and physical information. The project involves training selected components in an open‑loop setting and analyzing how World Model integration improves the geometric consistency and physical plausibility of AI‑based action or trajectory predictions, using offline datasets or logged data.

Key Research Areas

• Review of AI models for autonomous driving and their limitations in metric and geometric grounding
• Review of geometry‑aware World Models for environment representation
• Analysis of open‑loop failure cases caused by insufficient geometric information
• Design and training of a modular World Model + AI system with partial or lightweight learning
• Open‑loop evaluation using geometry‑aware metrics (e.g., distance consistency, collision risk, trajectory feasibility)
• (Optional) Lightweight integration or training strategies without large‑scale end‑to‑end retraining

Technical Prerequisites (or Motivation to Learn)

• Interest in autonomous driving, machine learning, or robotics
• Basic understanding of deep learning and model training
• Ability to read and modify research codebases
• Proficiency in Python
• Experience with offline datasets or representation learning is a plus

Kontakt: Christian.Prehofer@tum.de