Safe Reinforcement Learning for Autonomous Spacecraft Systems (m/f/d)
Applications are invited for a fully funded PhD position exploring the intersection of reinforcement learning, spacecraft dynamics, and formal verification.
Topics for Engineering Projects are published on this page. If you are interested in one of the topics, please write a short e-mail with 3-4 sentences to niklas.kochdumper@tum.de explaining why you are interested in the topic. For group applications, we also ask for a short description of the group's motivation for the topic. The topics will be assigned in mid/end of April (for the summer semester) or mid/end of October (for the winter semester). The submission of the report and the pitch for the engineering project are usually scheduled for the last week of lectures.
Eye in the Sky: Ceiling Camera Localization for Indoor Robotics
The goal of this project is to develop a ceiling-mounted camera system that acts as an indoor positioning tool for robots and objects on the floor. By calibrating the camera’s intrinsic parameters and estimating its pose relative to the environment, the system establishes a geometric mapping that converts image pixels into real-world floor coordinates. Students implement real-time object detection using color features or fiducial markers and use homography-based projection to determine each object’s position with high accuracy. The project offers hands-on experience in camera calibration, coordinate transforms, and real-time vision processing, and provides a practical platform for experiments in robot tracking and indoor localization.
This engineering project seeks one team of 2-3 students.
Lightweight Stereo Vision System for 3D Scene Reconstruction and Obstacle Abstraction
The goal of this project is to build a stereo vision system using Raspberry Pi-compatible IMX219 cameras to capture synchronized image pairs and compute depth maps. The resulting point clouds should then be processed into convex polytopes that abstract obstacles in the environment. These simplified models enable fast planning and mapping algorithms suitable for small robots and drones. The system is validated through controlled experiments involving obstacle detection and reconstruction.
This engineering project seeks one team of 2-3 students.
Preliminary Design of a Flying Spacecraft Simulator Based on Omnicopter Drones
Since omnicopter drones are capable of generating thrust in arbitrary directions, they are particularly well suited for compensating gravitational forces. The goal of this project is to equip such an omnicopter with reaction wheels and thrusters in order to simulate spacecraft motion under microgravity conditions. The key focus of the project is the proper dimensioning of both the reaction wheels and the thrusters, as well as the development of a corresponding CAD model for the complete spacecraft mockup.
This engineering project seeks one team of 2-3 students.