Master Thesis Projects 2027: Planning, Decision-Making and Safety - Cluster C

Help autonomous systems decide what to do next How should an autonomous vehicle act in a complex and uncertain world? At Zenseact, we're building systems that reason about risk, predict outcomes, plan safe behaviors, and make decisions in situations where safety matters. Cluster C focuses on the intelligence layer of autonomous driving, where AI moves beyond understanding the world to deciding how to act within it. As a thesis student, you'll work on research challenges spanning reinforcement learning, planning, verification, simulation, safety assessment, world models, and control systems. Your work will contribute to technologies designed to help autonomous systems make safer and more reliable decisions. Choose Your Project C1: Scenario Trajectory Parameterization Develop more efficient and meaningful representations of traffic trajectories and driving scenarios for analysis and planning. C2: Adaptive Stress Testing of Safety-Critical Automotive Software using Adversarial Reinforcement Learning Explore how reinforcement learning can discover challenging edge cases and safety-critical situations that traditional testing might miss. C3: Reasoning / QA Training of VLA Models for Parking Investigate reasoning capabilities in modern Vision-Language-Action models and their application to autonomous parking systems. C4: Object-Level Trajectory Planning Using World Models Research how world models can support better decision-making and trajectory planning in complex driving environments. C5: Virtual Verification First for E2E AD Software Development Framework Study how simulation and virtual verification can improve the development and validation of autonomous driving software. C6: Clustering Methods for Cut-In Scenarios Develop techniques to organize, categorize, and understand challenging traffic interactions involving cut-in behavior. C7: Evaluation of Predictors for Safety Assessment Investigate methods for assessing how predictive models contribute to safer autonomous driving decisions. C8: Neural Controller for Parking Scenarios (This project is in Lund, Sweden) Explore neural-network-based control systems for automated parking applications. C9: Evaluation of Search-Based Falsification Methods for Safety-Critical Driving Software Research methods for systematically uncovering weaknesses and failure modes in autonomous driving software. Read more about each project here: Cluster C Your Mission Every Cluster C thesis combines research with real-world experimentation. You will work with autonomous-driving datasets, simulation environments, and modern AI methods to investigate how intelligent systems reason, plan, and make decisions under uncertainty. Depending on your project, you may: - Develop and evaluate algorithms - Work with simulation and virtual testing environments - Design experiments and safety metrics - Investigate edge cases and failure scenarios - Analyze planning, prediction, or controller behavior - Evaluate robustness, performance, and safety - Present findings and recommendations You'll have significant ownership of your work while receiving guidance from experienced engineers and researchers. Who You Are We're looking for curious students who enjoy solving difficult problems and exploring new ideas. You're currently pursuing a Master's degree in Computer Science, AI, Machine Learning, Robotics, Electrical Engineering, Engineering Physics, Applied Mathematics, or a related field. We value different perspectives, backgrounds, and ways of approaching problems. You don't need to meet every preferred qualification to apply. Strong foundations, curiosity, and a willingness to learn matter most. Essential What you bring: - Programming experience, preferably in Python - An interest in machine learning, planning, optimization, simulation, control, or safety - Ability to learn from and evaluate scientific research - Ability to communicate technical ideas and findings in English Helpful, but Not Required Experience in machine learning, reinforcement learning, planning, robotics, simulation, safety-critical systems, or research-driven software development will be valuable. This may come from coursework, research projects, internships, open-source contributions, or personal projects. Why Cluster C? Cluster C explores some of the most challenging questions in modern AI: - How should an intelligent