Preliminary posting. This position still has to be confirmed by the Intelligent Vehicles Lab and by HR; details may change.
Today’s automated vehicles are very good at the situations they have seen and fragile at the ones they have not. Generative models change that picture: world models that imagine how a scene will unfold, vision-language-action models that turn perception into decisions and can say why, and foundation models trained on far more driving than any single fleet records. The open question is how far this carries — from narrow, task-specific autonomy towards a driving intelligence that generalises across cities, sensors and the long tail of rare events. This PhD works on exactly that question, on our own research vehicle and in closed-loop simulation.
Your research
- Develop generative world models and vision-language-action models that predict how a traffic scene evolves and plan the vehicle’s behaviour in it, with vulnerable road users at the centre.
- Investigate what makes these models generalise: scaling data and model size, self-supervised pre-training on unlabelled drives, map and scene representations, and reasoning in language.
- Make the decisions inspectable: explanations that match what the model actually used, and evaluation that goes beyond open-loop metrics.
- Evaluate in closed loop — in CARLA and on benchmarks such as HABIT — and bring the best models onto the lab’s research vehicle.
- Publish at leading venues (e.g. CVPR, ICCV/ECCV, NeurIPS, ICLR, IROS, IEEE IV) and release code and data.
Your tasks
- Research, implementation and experiments towards your doctoral thesis.
- Work with the PhD candidates on BEVDriver, BEV-LLM and the lab’s generative-planning line, and with partners in publicly funded projects.
- Co-supervise Bachelor and Master theses and support teaching in machine learning and autonomous driving.
Your profile
- An excellent Master’s degree in computer science, electrical engineering, robotics, mathematics or a related subject.
- Solid deep-learning background; hands-on experience with transformers, diffusion models or large (vision-)language models.
- Strong Python and PyTorch; experience with multi-GPU training is a plus.
- Interest in autonomous driving; experience with CARLA, nuScenes/Waymo data or ROS 2 is welcome but not required.
- Very good English; German is helpful but not required to start.
What we offer
- A research topic at the front of the field, with a research vehicle, a driving simulator and GPU compute to work with.
- A small, collaborative team and close supervision by the lab head.
- Funded conference travel and the expectation that you publish.
- A doctorate through a Bavarian doctoral centre for universities of applied sciences, together with a partner university.
- Employment under TV-L with the benefits of the public sector, on the Lothstraße campus in central Munich.
How to apply
Send one PDF to e-mail (enable JavaScript) with “IVL-2026-PHD-GEN” in the subject: a one-page motivation letter (why this topic, why this lab), your CV, degree certificates and transcripts, a piece of writing (Master’s thesis or paper) and the names of two referees. Applications are reviewed on a rolling basis until the position is filled.
Equal opportunity
Hochschule München promotes equal opportunity and welcomes applications from women. Applicants with severe disabilities are given preference where qualifications are equal.