My research focuses on the intersection of computer vision, machine learning, and control theory. On one side, I am interested in representations that enable the efficient combination of scene understanding and decision-making algorithms. These can vary from bird's eye view scene and map understanding in autonomous driving to using NeRFs for active 3D reconstruction. On the other side, I am interested in learning safe multi-agent policies directly from data, using principles from control and game theory to guide the learning process. Such methods can, for example, directly learn driving policies from human driving data and can be used for motion prediction, interactive motion planning, or data-driven simulation. The combination of better representations and policy learning will allow us to develop models that pave the path for truly autonomous robotic systems.