Student Research Programs

Sponsors in 2025 include the following:

• Sponsor: DB InfraGO AG
  • Project Title: Learning to create railway disposition concepts
    Delays, disturbances and disruptions are the daily business of railway infrastructure managers and operators. While modern optimization methods are well-developed for strategic and operational planning tasks, real-time management often lies solely in the hand of experienced dispatchers. In the MODAL MobilityLab, we want to bridge this gap. A simple idea to deal with the time criticality is to pre-compute disposition concepts. This allows in principle to apply mathematical models that are targeted at planning, but those are typically computationally challenging, and they do not necessarily follow the typical disposition philosophy. Instead of a global system optimum, dispatchers prefer to follow a concept that sticks as much as possible to the regular timetable, and differs only in a very limited number of places. It is therefore natural to think about disposition actions, such as short-turning and changing train orders, track allocations, or dwell times, and to create an optimized disposition concept that applies a selection of these actions. The challenge is to find out which of these actions are necessary, impactful, and improving. Speaking in mathematical terms, we want to find good quality solutions to a hard combinatorial optimization problem by using a multi-agent reinforcement learning approach. The project targets at first at understanding how to model the railway disposition concept optimization problem in terms of mixed-integer linear programming, then to define suitable agents and their actions, and finally to implement and evaluate learning strategies on data provided by DB InfraGO AG, Germany’s largest railway infrastructure manager.We particularly welcome students with a background in discrete optimization or machine learning and expect familiarity with Python.The project is supervised by MODAL Mobility Lab in collaboration with industry partner DB InfraGO AG.

• Sponsor: 1000shapes GmbH
  • Project Title: Topology-Aware Non-Rigid Registration for Medical and Vision Applications
    Non-rigid registration under topological changes is an important area of research in computer vision and graphics that finds applications in diverse areas such as registering a template of a bone to radiographs of fractures, using pre-operative CT models to aid surgery with Augmented Reality (AR) during organ resections, tracking glacier calving from satellite images, tracking mitosis and meiosis of cells from microscopic images, using food processing robot for slicing meat and fresh produce, to name a few. For such a non-rigid registration under topological changes, given a template shape denoted by an atlas of some charts, the key challenge is to: a) determine a suitable shape signature that can be used to match the template with pointcloud, b) find a strategy for determining optimal separation boundaries in the charts and c) obtain a provably convergent optimization that solves the shape to pointcloud registration problem. We hope to solve the problem of non-rigid registration under topological changes with the following steps: 1) establishing an initial correspondence (between template and observed data) with shape signature matching, 2) imparting a graph structure to the template and observed pointcloud so that we may minimize an alignment cost while determining an optimal graph cut that splits the shape to best explain the observation. The mathematical techniques required for achieving such a methodology involve non-convex optimization, Semi-Definite Programming (SDP), fundamentals of machine learning and preferably, some understanding of differential geometry.The project is supervised by MODAL Med Lab in collaboration with industry partner 1000shapes GmbH.

• Sponsor: FICO Optimization
  • Project Title: Matrix-Inspection-Tool for Mixed-Integer Programming
    The project aims to develop a versatile tool in Python designed to enhance the analysis and understanding of Mixed-Integer Programming (MIP) problems by analyzing the matrices that define these problems. The tool will provide various statistics and visualizations for each problem. It will be released under the MIT license, allowing it to be freely used and modified by the community. The project requires knowledge of Python and basic statistics. Students with an interest in mathematical optimization will find this project particularly appealing. The project is supervised by MODAL SynLab in collaboration with industry partner FICO , which provides market-leading optimization solvers for mixed-integer programming.