28 DAILY WACV Congrats, Doctor Katrin! Sunday Katrin Renz has recently completed her PhD at the University of Tübingen and the Max Planck Institute for Intelligent Systems (IMPRS-IS). Supervised by Andreas Geiger, her research focuses on end-to-end autonomous driving, specifically combining vision, language, and action to make self-driving cars smarter and more interpretable. Katrin is currently building her own startup, continuing her work on embodied AI. Congrats, Doctor Katrin! Autonomous driving promises to make roads safer, but building systems that can handle the full complexity of real-world traffic remains a major challenge. In particular, existing systems struggle to generalize to the "long tail" of rare or unusual driving scenarios. These infrequent but safety-critical edge cases are essential to master for reliable, real-world deployment. Katrin’s thesis tackles this gap by leveraging the broad world knowledge of foundation models and grounding their reasoning power in the physical world. Her goal? To teach cars not just to drive safely, but to reason, explain their actions, and interact in natural language. In her first major work, Katrin introduces PlanT, a transformer-based planner. Moving away from computationally heavy, pixel-level Bird’s-Eye View (BEV) images, PlanT uses a compact object-level representation. By leveraging a standard transformer architecture inspired by language modeling, it achieves expert-level driving performance on the CARLA simulator. Best of all, PlanT’s attention weights make the decision-making process highly explainable, clearly highlighting the most relevant objects in a traffic scene without requiring extra manual annotations. Her second project, DriveLM, directly integrates language as an additional modality, making it one of the pioneering works to explore VisionLanguage-Action (VLA) models for autonomous driving. By bringing these components together into a unified architecture, she formulates driving as
RkJQdWJsaXNoZXIy NTc3NzU=