Available Masters' Theses

The goal of this thesis is to develop an interactive front-end tool capable of visualizing the effects of traffic policies. At its core, the tool leverages a Graph Neural Network (GNN)-based surrogate model that enables rapid simulations. The aim is to provide decision-makers and citizens with an intuitive platform to test various traffic measures in different urban areas and scales. The impacts of these measures should be displayed accurately and in a visually appealing way. Tasks include selecting a suitable framework (e.g., React, Dash, or Streamlit), seamless integration of the surrogate model, and developing user-friendly visualizations. The tool should be fully operational and ready for practical use. Strong programming skills are a prerequisite for this thesis.

This thesis explores how transfer learning can be utilized to adapt a Graph Neural Network-based surrogate model for evaluating traffic measures in one city to other cities. Using simulation data from multiple cities, various transfer learning techniques such as graph-level transfer and multi-task learning will be applied and compared. The goal is to develop models that can learn generalizable urban patterns, enable robust predictions for new cities, and account for data heterogeneity. The results will be validated using simulated data. Strong programming skills are a prerequisite for this thesis.

This thesis focuses on recognizing the intentions of extra vulnerable road users (eVRU), specifically wheelchair users. The analysis relies on camera and Lidar data, utilizing both conventional algorithms and deep learning approaches.

This thesis focuses on recognizing the intentions of vulnerable road users (VRU), specifically pedestrians and cyclists. The analysis relies on camera and Lidar data, utilizing both conventional algorithms and deep learning approaches.

In the bicycle simulator, the choice of the ridden path and the speed can be decoupled. In this study, we investigate the ridden path and focal points in driving simulator studies in order to use them for microscopic modeling of cyclists.

This thesis focuses on recognizing the intentions of cyclists at an unsignalized intersection. The analysis relies on camera data, which is already available from a previous bike simulator study. The goal is to find out the importance or relation between body gestures or movements to the final maneuver decision, utilizing both conventional algorithms and deep learning approaches.

This thesis focuses on analyzing the cyclists behavior at dynamic shuttle stops. The analysis relies on both trajectory and the camera data of the cyclitsts, which is already available from a previous bike simulator study. The goal is to find out the correlation between trajectory and body gestures and body movements of cyclists.

The aim of the thesis is to study the routing of a fleet of vehicles (with big batteries) that can charge other on-demand ride-hailing or ride-pooling vehicles (with smaller batteries). The thesis will first research the available methods for the routing of on-demand charging vehicles, develop a new routing scheme and then evaluate their efficiency in an agent-based simulation framework, called FleetPy.

The aim of the thesis is to study the routing of a fleet of vehicles (with big batteries) that can charge other privately owned electric vehicles (with smaller batteries). The thesis will first research the available methods for the routing of on-demand charging vehicles, develop a new routing scheme and then evaluate their efficiency in an agent-based simulation framework, called FleetPy.

In automated ride-pooling services, trip requests are dynamically processed by a central optimizer to assign new routes to fleet vehicles and serve customer requests. Once a vehicle has completed the route, however, the question arises as to where it should wait for new assignments. Various strategies are conceivable: The search for the next parking lot, or a return to the depot. The aim of the thesis is to work out different strategies and to implement and evaluate them in a simulation. Within the work, the strategies are to be implemented in a framework developed at the chair, consisting of FleetPy and SUMO, and evaluated for operational and traffic effects.

The operational problem of ride-pooling is very challenging because of its dynamic and stochastic nature. Based on estimations of future demand, operators can reposition their (idle) vehicles to be prepared better for the future. Since the rewards of making these decisions evolve over time, creating an objective function for repositioning is non-trivial. This is where machine-learning methods (e.g. value function estimation or reinforcement learning) can come into play. With FleetPy, a simulator will be available, which should be extended in this thesis

The goal of the thesis is to analyze how different service designs, network structures and demand patterns, which relate to transfers between line-based public transport and a ridepooling service, affect the performance indicators of a ridepooling fleet. The chair’s fleet simulator FleetPy can be used as basis to model a ridepooling service. A small programming extension is necessary to model it as access and egress service for public transport.

In ride-pooling services, customers book a trip with the provider via an app and the trips are dynamically integrated into the route planning of fleet vehicles. No-shows of a booked trip can result in costs for the operator as well as for other customers. The goal of this work is to integrate no-shows into the existing simulation environment "FleetPy" and to simulate the effects on the overall system.

Research question: how to choose a right map matching algorithm balancing accuracy and computational time? By comparing the performances of different map-matching algorithms, the student is expected to find a balance in the accuracy of map-matching of GPS trajectory data to network graph for a dynamic and static use case within a reasonable computational time.

This thesis explores integrating autonomous vehicle navigation through urban construction sites into existing path planning algorithms, focusing on adapting for safe passage. The evaluation of the adapted planning behavior is conducted and assessed within a simulator, forming a vital part of the research.

This thesis focuses on integrating dynamic stops into the route planning of automated shuttle buses, aiming to efficiently adapt existing planning systems for urban transport networks. The evaluation of the planning behavior is conducted and assessed in a simulator, constituting a central part of the study.

This thesis explores Facial Emotion Recognition through Convolutional Neural Networks, aiming to advance human-computer interaction by accurately identifying human emotions from facial expressions.

This project focuses on integrating e-scooter simulation within Unity and VR environments, employing an actual electric scooter for immersive and realistic user experiences. It aims to bridge the gap between virtual simulations and real-world scooter maneuvering, enhancing training and entertainment applications.

In many large-scale sporting events, such as cross-country skiing marathons, bottlenecks and overcrowding result in the formation of congestion analogous to that observed in road traffic. Using publicly available split times as stationary detector data and GPS tracks as floating car data, the goal of this thesis is to perform a detailed analysis of the congestion formation in several high-profile cross-country ski marathon events.

The accurate estimation of queue lengths at traffic lights is a critical point in order to efficiently adapt the control system. An existing AI method based on drone data is to be tested and further developed as part of this project. The existing code and approaches for improvement will be provided by the mentor.

This thesis explores the use of AI techniques to convert traffic safety rules from human language into machine-readable formats, enhancing the ability of autonomous vehicles to understand and adhere to safety guidelines. By employing natural language processing (NLP) and formal logic-based models, this work aims to bridge the gap between human-expressed regulations and machine interpretation, facilitating precise adherence to complex traffic rules. This formalization process is crucial for enabling autonomous systems to exhibit compliant and predictable behaviors across varying traffic scenarios, contributing to safer and more reliable transportation systems.

This thesis will explore human driver behavior in roundabouts by analyzing trajectory data from multiple locations. The student should perform data analysis to identify patterns and later use machine learning to classify driving behaviors and develop a model to predict driver intentions.

Artificial intelligence methods are currently driving the development of automated driving. This thesis focuses on how the interaction between automated vehicles (and other road users (optionally)) can be optimized with the help of artificial intelligence. Existing conventions on right of way or fixed lane allocation may be challenged.

Similar to lane-based traffic, where vehicles choose a proper lane upon an intersection, in the lane-free case, the vehicle should also adjust their lateral location properly before reaching an intersection depending on their turn movement. This work develops optimal approaches for choosing lateral locations for vehicles.

Lane-free traffic allows connected and automated vehicles to move across the entire road surface in a flexible fashion. This allows for a reorganization of today's traffic control at intersections. To conduct an evaluation of the developed ideas, a self-developed simplified simulation environment (e.g. Cellular Automata) can be used.

With the introduction of Connected Autonomous Vehicles (CAVs), there is a growing interest in Lane-Free Traffic (LFT), where CAVs drive without managed lanes. The movement of CAVs in LFT is controlled by control algorithms that coordinate their movements. However, the performance of this controller can be severely affected by the presence of human drivers. Therefore, this thesis develops a control strategy for CAVs capable of dealing with mixed traffic where humans and CAVs share the same road without degrading the performance of CAVs in LFT. The simulation will be conducted in a customized SUMO for LFT and requires some basic programming skills in C++.

The openPASS platform has recently been developed by the BMW Group to conduct scenario-based traffic simulation. Compared to other standard traffic simulators, advanced driver assistance and automated driving systems are integrated into openPASS, allowing for the advanced simulation of vehicle-based scenarios. The goal of this thesis is to understand how openPASS works, what its capabilities are, and how to develop Traffic simulation of highway, rural, and urban scenarios.

This thesis aims to utilize and evaluate various causal inference approaches to assess the impacts of traffic measures. The goal is to establish a theoretical foundation to determine the causal effects of specific interventions, such as the introduction of new modes of transportation or changes to infrastructure, on key traffic outcomes. The application will be built on a Graph Neural Network (GNN)-based surrogate model.

This thesis explores the implementation of a two-directional roundabout in SUMO with adaptive lane configurations for the two directions based on real-time traffic demand. The student will develop a dynamic lane allocation system, fine-tune its parameters, and analyze its potential for different roundabout configurations.