Student Projects

Type of Project: Semester Project
Supervisor: Yasaman Zolfimoselo ([email protected])
Student: Open to application

Multi-UAV systems offer a flexible alternative to fixed sensing infrastructure for urban traffic monitoring. In this project, a drone fleet monitors an urban road network and must coordinate the trajectories of its UAVs over a finite horizon to prioritize high-importance roads. The reward is history-dependent: the value of re-observing a road recovers with the time since its last visit, so the team reward depends on the entire joint visit history rather than the current joint state. The dependence on past visits makes standard Markovian RL approaches difficult to scale. This project will investigate decentralized recurrent policies that allow each drone to use information from its local observation history when coordinating its monitoring decisions.

Requirements: 
Strong background in reinforcement learning and machine learning.
Solid Python programming skills; experience with NumPy and PyTorch required.

References: 
Maljkovic, M. and Geroliminis, N., 2025, June. On Learning-Based Traffic Monitoring With a Swarm of Drones. In 2025 European Control Conference (ECC) (pp. 1842-1847). IEEE.
Cayci, S. and Eryilmaz, A., Recurrent Natural Policy Gradient for POMDPs. Transactions on Machine Learning Research.

Contact: Please email a short paragraph describing your background and motivation, along with your CV and academic transcripts, to [email protected]

Type of Project: Master Project
Supervisor: Manos Barmpounakis ([email protected])
Student: Open to application

This project aims to investigate how the composition of traffic (e.g., cars, motorcycles, buses and trucks) influences the formation and evolution of congestion in urban environments.

The student will use Python to process and analyze high-resolution vehicle trajectories extracted from drone recordings and derive traffic indicators such as speed, density, flow, vehicle composition, and congestion levels. The project will explore whether information about the mix of vehicles on the road can help explain and predict traffic congestion and whether different traffic compositions lead to different congestion and recovery patterns (hysteresis loops).

The topic is intentionally broad and can be adapted according to the student’s interests and background. The project is data-oriented and computationally intensive, and therefore requires strong Python programming skills, particularly for handling large datasets, data analysis, and visualization.

Type of Project: Bachelor Project / Master Project
Supervisor: Can Chen ([email protected]); Ran Chen ([email protected])
Student: Open to application

The imminent penetration of low-altitude passenger and delivery aircraft into the urban airspace will give rise to new urban air transport systems, which we call low-altitude aircraft transport (LAAT) systems. Emerging LAAT systems involve largescale point-to-point operations, which require new management schemes distinct from the centralized flight control used in conventional aviation. Macroscopic Fundamental Diagrams (MFDs) have been widely investigated for urban traffic management. Recent advances demonstrate that MFD-based departure management and boundary control are promising ways to improve urban air mobility. However, few existing MFD-based traffic control approaches have adequately addressed LAAT systems with prioritized aircraft (PA), which in reality includes but not limited to aircraft that carry more passengers or more valuable goods such as emergency relief supplies.

A key aspect of this project is the incorporation of airspace separation, allowing for dedicated regions for PA in certain airspace and mixed traffic operations in others. In this project, you will learn how to model LAAT systems within the MFD framework and how to devise efficient boundary control and departure management strategies for optimizing the urban air mobility.

This project is designed for two Bachelor students or one Master student. We welcome candidates with a keen interest in aggregate traffic modeling and Macroscopic Fundamental Diagrams (MFD), as well as a strong background in traffic flow theory, systems and control, and linear algebra. While MATLAB is the primary programming tool for this project, students proficient in Python are also encouraged to apply.

Type of Project: Bachelor Project / Master Project
Supervisor: Marko Susnjar ([email protected])
Student: Open to application

Max-Pressure traffic signal control relies on up-to-date information about queue lengths at signalized intersections in order to select efficient signal phases. In practice, obtaining fresh queue measurements at every intersection would require dense fixed sensing infrastructure. This project investigates how mobile sensing can support Max-Pressure control by using a limited fleet of drones to dynamically visit intersections where updated information is most valuable for traffic control.

Students will develop strategies for dynamically selecting which intersections should be observed. Selection may depend for example on estimated congestion, queue ratios between competing phases, or the time since an intersection was last monitored. These adaptive strategies will be compared with a fixed sensing configuration, where the same intersections are selected at the beginning of the simulation and remain monitored throughout. The project will use either an existing MATLAB traffic simulator or a new simulation environment based on SUMO and Python.

The goal is to determine whether adaptive sensing can improve traffic performance while observing only part of the network, and how much sensing coverage is needed to approach the performance of a fully observed system.

Type of Project: Bachelor Project / Master Project
Supervisor: Marko Susnjar ([email protected])
Student: Open to application

Many traffic-control strategies rely on aggregated traffic information rather than measurements from every road. For perimeter control, important quantities include the number of vehicles within a region, flows between neighbouring regions, and Macroscopic Fundamental Diagrams. In this project, we investigate whether drones can provide enough local measurements to estimate such regional traffic states in real time despite observing only a small fraction of the network.

Students will use complete traffic data to calculate ground-truth regional quantities and then restrict the available information to sparse observations representing a small drone fleet. They will develop and compare simple estimation methods, such as scaling, linear regression, Random Forests, or other suitable machine-learning approaches. The project will study how estimation accuracy changes with the number and location of observations. It requires working in Python and either basic knowledge of machine learning or motivation to learn introductory ML methods.

The goal is to identify sensing and estimation approaches that can provide reliable regional traffic information for future perimeter-control applications. An optional extension is to reconstruct a Macroscopic Fundamental Diagram from sparse observations.

Type of Project: Bachelor Project / Master Project
Supervisor: Meng Xu ([email protected])
Student: Open to application

This project studies persistent monitoring with multiple UAVs in a grid-based environment. The main goal is to understand how UAVs should coordinate their movements to maintain fresh information across the monitored area. Depending on the student’s interests, different approaches can be explored, including control-based methods, heuristic strategies, or reinforcement learning. Possible research questions include how fleet size affects monitoring efficiency, whether decentralized UAVs naturally form stable monitoring territories, and how UAVs should balance frequently monitoring important areas against revisiting areas whose information has become outdated. The project will use simulation to compare different strategies in terms of coverage, information freshness, redundancy, and coordination efficiency. Basic knowledge of Python programming is required for implementing the simulation and algorithms.

Type of Project: Semester Project
Supervisor: Weijiang Xiong ([email protected])
Student: Zhixiang Dai

Urban traffic is highly periodic due to the regular activity patterns of residents, and at the same time it contains significant stochasticity coming from the complex interactions. In the literature, this phenomemon has been implicitly modeled, i.e., the neural network is trained to just work with the observed traffic statistics. However, in this project, we aim to decompose the periodic part and the stochastic part by training a neural network to predict the residual (actual traffic – historical average). At the same time, the input will be explicitly augmented memory of the past traffic (all historical traffic data).

For a smooth progress of the project, the student is expected to have some knowledge on deep learning and preferably hands-on experiences with PyTorch.

Reference: https://github.com/Jimmy-7664/ST-SSDL

Type of Project: Semester Project
Supervisor: Marko Susnjar ([email protected])
Student: Arthus Duval, Nicolas Künzli

Traffic congestion is a persistent challenge in many cities worldwide, especially during peak hours. Congestion often begins at specific bottlenecks and then propagates to neighboring streets, creating complex and evolving traffic patterns. Many traffic control algorithms divide a city into fixed and predefined regions, which cannot capture the dynamic nature of urban traffic. Hence, proper identification of congested and uncongested regions, and their evolution in the real-time, can be beneficial for traffic control.

In this project, students will apply mathematical methods, such as Support Vector Machines, to classify city streets (modelled as graph edges) as either congested or uncongested, using real-world traffic data. The classification result will then be used to identify main pockets of congestion in the city and divide the area into clusters. The goal is to create nearly homogenous clusters of neighbouring streets based on congestion, and to analyse how cluster boundaries evolve over time. These results will be compared with common clustering methods, such as K-means clustering.

A solid foundation in mathematics and strong programming skills with experience in Python/MATLAB are preferable.

Type of Project: Semester Project
Supervisor: Yasaman Zolfimoselo ([email protected])
Student: Emile Matar

Multi-UAV systems offer highly effective solutions for dynamic urban traffic monitoring. In this project, we focus on the cooperative optimization of a drone fleet tasked with efficient monitoring of a road network, with the goal of coordinating UAVs’ trajectories so that road segments with higher monitoring importance are prioritized.

The student will work with a grid-based simulation of an urban road network. In this model, the importance value of each grid cell is dynamic, evolving based on the drones’ observation history. The challenge is to design a tractable decision-making algorithm (centralized or decentralized), based on dynamic programming principles, that can optimize the fleet’s objective over time.

Solid programming knowledge in Python is required. Familiarity with Dynamic Programming or Reinforcement Learning is an advantage.

Type of Project: Semester Project
Supervisor: Batuhan Avci ([email protected])
Student: Muratcan Akgün, Mahdi Fourati

Urban traffic congestion can be modeled as a function defined on a road-network graph. Each node represents a location such as a road segment, and edges encode connectivity in the network. A Gaussian process can act as a probabilistic function approximator on this graph, providing predictions together with uncertainty estimates.

This semester project focuses on Gaussian processes defined over graph nodes and implements graph-based kernels to interpolate traffic values from a limited set of observed sensors to all nodes in the network. Given partial observations, the model should infer the latent traffic field over all nodes and output both the interpolated congestion values and the corresponding posterior uncertainty.

A case study will use real traffic data collected by drones.

Required skills: Python; basic probability and statistics; linear algebra; familiarity with machine learning fundamentals and non-parametric learning.

Nice-to-have / can be learned during the project: Gaussian processes; basic graph concepts; experience with ML libraries/frameworks (e.g., PyTorch, GPyTorch).

Type of Project: Semester Project
Supervisor: Meng Xu ([email protected])
Student: Eugénie Hoelzl, Arthur Van den Broeck

Urban bus corridors often suffer from localized bottlenecks (e.g., a highly congested stop or junction). Delays formed at such critical points can propagate along the whole line, causing unreliable headways, passenger crowding, and long waiting times. Traditional fixed-route operations have limited flexibility to absorb these disruptions.

This project studies a coordinated transit system that integrates (i) skip-stop control for fixed-route buses at a congested area and (ii) a small fleet of on-demand minibuses that dynamically supports regular bus services. The key practical challenge lies in the online joint optimization of when to admit requests, which passengers to admit, and which minibus should serve them, while guaranteeing feasibility for mandatory tasks triggered by skip-stop operations.

Depending on your background and the thesis timeline, The student will contribute to one or more of the following:
1. Improve the lower-level routing algorithm (quality, robustness, runtime).
2. Design and implement the upper-level admission–assignment controller (optimization or other suitable methods).

We welcome students with interest/skills in at least one of the following:
1. Python programming (required);
2. simulation-based optimization, heuristic routing, mixed integer linear programming
3. learning-based optimization methods

Type of Project: Semester Project
Supervisor: Weijiang Xiong ([email protected])
Student: Zhiyan Ke

In this project we aim to learn the vehicle motion dynamics using massive trajectories extracted from high-quality drone videos taken at Songdo, Korea [1]. We will process the data in accordance with the UniTraj framework [2] and benchmark the performance of current vehicle trajectory models in forecasting and generation tasks. The project can start with a smaller dataset SinD [3] and then move on to the Songdo dataset.

Reference:
[1] https://www.sciencedirect.com/science/article/pii/S0968090X25002098
[2] https://github.com/vita-epfl/UniTraj
[3] SinD dataset: https://arxiv.org/pdf/2209.02297

Type of Project: To be discussed with the student
Supervisor: Weijiang Xiong ([email protected])
Student: Open for application

Learning the pattern of traffic data is a fundamental task in transportation systems, but most existing methods only provide deterministic predictions, which can neglect the inherent stochasticity in urban traffic. Therefore, this project aims to apply the concept of diffusion models (DDPM) [1] into traffic forecasting and enable probabilistic predictions.

The project can start with a closely related work [2] and transfer the model to two widely acknowledged datasets METR-LA and PEMS-Bay. Depending on the progress, the method can be extended to traffic data imputation, i.e., missing value completion. For a smooth progress of the project, the student is expected to have some knowledge on deep learning and preferably hands-on experiences with PyTorch. The theoretical part of DDPM can be difficult to start with, but it is not mandatory for the success of this project.

[1] Denoising Diffusion Probabilistic Models https://arxiv.org/pdf/2006.11239
[2] DiffSTG: Probabilistic Spatio-Temporal Graph Forecasting with Denoising Diffusion Models https://dl.acm.org/doi/pdf/10.1145/3589132.3625614

Type of Project: To be discussed with the student
Supervisor: Weijiang Xiong ([email protected])
Student: Open for application

Existing traffic prediction methods are usually dataset-specific, which limit their ability to generalize across different cities. This project aims to explore possible solutions to more general traffic prediction models. Concretely, we will apply the idea of MAE [1] into the traffic speed datasets METR-LA and PEMS-Bay. We will then fine-tune the pretrained model on a new urban traffic dataset and verify the effectiveness and benefit of the pretraining step. Alternatively, we can start with the model in OpenCity [2]. An ideal student should have some hands-on experiences about deep learning.

[1] Masked Autoencoders Are Scalable Vision Learners https://arxiv.org/abs/2111.06377
[2] https://github.com/HKUDS/OpenCity

Type of Project: To be discussed with the student
Supervisor: Ran Chen ([email protected]); Can Chen ([email protected])
Student: Open for application

The emergence of Low-Altitude Air City Transport (LAAT) systems, involves piloted and autonomous drones for passenger and cargo transport. The inevitable penetration of drones into urban airspace necessitates robust traffic management tools will fundamentally transform the futrue of urban mobility.

One existing multi-drone simulator (see https://doi.org/10.1016/j.trc.2023.104141) applies the Artificial Potential Field (APF, see http://dx.doi.org/10.1109/tits.2021.3096558) for collision avoidance. While simple to implement and generally effective, this approach has two critical limitations that impact the accuracy of our simulations, particularly in congested airspace. First, the PF method does not guarantee collision avoidance, meaning drones can still crash into one another. Second, it lacks the ability to enforce flight boundaries, which can cause drones to stray from their designated regions. These issues can skew simulation results, leading to an inaccurate MFD and compromising the integrity of our statistical analysis.

Therefore, this two-student project aims to address these limitations by evaluating the current APF algorithm (including parameter sensitivity) and researching novel, provably safe collision avoidance methods (e.g., based on optimization, geometric, or control-theoretic approaches). The core task involves implementing and comparing one or more new algorithms against the existing PF approach within the simulator. The final deliverables include constructing the LAAT MFD corresponding to the updated algorithms and rigorously evaluating performance, which may include congestion reduction, safety guarantee, etc.

We seek students proficient in Control Theory, Motion Planning, and MATLAB.

Type of Project: To be discussed with the student
Supervisor: Yura Tak ([email protected])
Student: Open for application

This project proposes an innovative approach that leverages drone-captured aerial videos to monitor parking areas and extract key transport indicators related to parking availability and usage patterns. Using computer vision and video analytics, the system will detect parked and moving vehicles, estimate occupancy levels in parking zones, and generate insights such as percentage of available vs. occupied spaces, temporal demand trends, spatial heatmaps of high-occupancy areas, and short-term predictive models of parking availability. The project will deliver a parking occupancy estimation pipeline, a visualization dashboard for real-time and predictive indicators, and a case study analysis of parking demand.

The student should have a good programming skills in Python. Previous experience with machine learning and computer vision is a plus.

Type of Project: To be discussed with the student
Supervisor: Meng Xu ([email protected])
Student: Open for application

Unmanned Aerial Vehicles (UAVs) are playing an increasingly critical role in modern urban systems, especially in logistics and infrastructure monitoring. This semester project aims to explore intelligent UAV planning strategies that jointly handle parcel delivery and road surveillance tasks in a unified urban environment.

The student will work with a graph-based model of a real-world road network, where each road segment has a dynamically evolving “importance score” that increases with drone visits (i.e., surveillance) and decays over time if left unobserved. The challenge is to design a decision-making algorithm that balances two conflicting objectives: fast and efficient delivery vs. wide and regular network monitoring.

Solid programming skills in Python is a must.

Type of Project: Bachelor Project / Master Semester Project
Supervisor: Georg Anagnostopoulos ([email protected])
Student: Alia Boughaleb; Aurora Villain; Loic Antille; Soorya Pasupathy

Micromobility is becoming increasingly prevalent in cities. As a consequence, microvehicles, such as motorcycles, scooters and bikes, often share the same road space with conventional cars. We will refer to this coupling of a predominantly lane-based host system with a layer of lane-free parasitic flows as multispecies traffic. There are good reasons to believe that multispecies traffic is macroscopically a porous flow, where microvehicles may bypass congestion by percolating through the car traffic exactly lilke water percolating a porous substrate. In other words, multispecies traffic is an example of directed percolation (DP). DP entails the existence of a nontrivial, far from equilibrium phase transition from a disordered subcritical phase to an ordered supercritical phase that depends on a temperature-like control parameter and is governed by a universal power law relation. We argue that the temperature in our model is connected in a subtle way with the noise in the Vicsek model of collective motion [1].

In this project the students will be exposed to simulation techniques, as well as to more advanced concepts from complexity science, such as universality, criticality, ordering and (nonequilibrium) phase-transitions. Our focus is to address the research question of how, as we approach the so-called thermodynamic limit, the apparently chaotic dynamics of multispecies traffic comply to an underlying order, which can be precisely captured by a single, universal number known as the critical exponent. Critical exponents matter because they are surpisingly invariant to different system configurations and are also independent of the specific problem formulation.

References: [1] T Vicsek, A Czirók, E Ben-Jacob, I Cohen, O Shochet, Novel type of phase transition in a system of self-driven particles. Phys. Rev. Lett. 75, 1226–1229 (1995).

Type of Project: Master Semester Project
Supervisor: Minru Wang ([email protected])
Student: Jingren Tang; Lucas Maneff

Drones can be used to deliver packages in an urban environment. To curb excessive drone flight distances, a drone delivery system can be designed to transport parcels using unexplored capacities on scheduled public transit lines. The objective of this project is to explore matching heuristics that focus on the strategic selection of bus stops, implement proposed strategies in a simulation, and evaluate system performance.

Experience with Python and simulations is necessary. Knowledge on classical Vehicle Routing Problems and related problems is an asset, but not mandatory. The student(s) will get a chance to learn about Pick-Up and Delivery Problems and related optimization tools in Python.

Type of Project: Bachelor Project
Supervisor: Weijiang Xiong ([email protected])
Student: Chaimaa Ouchicha; Lise Gillin

The traffic forecasting problem is usually treated as a single-variate problem, where the forecasting methods concerns only one traffic variable, e.g., traffic speed or traffic flow. This project aims to develop a joint prediction approach for urban traffic speed, flow and density at the same time. Since these variables are gonverned by physical rules, this project also seeks to integrate the training of a prediction model with the physical relationship.

Type of Project: Bachelor Project
Supervisor: Weijiang Xiong ([email protected])
Student: Andrea S. F. Montoneri

Most existing traffic forecasting models predicts a deterministic estimation for future traffic, which can neglect the inherent stochasticity of urban traffic. Therefore this topics aims to develop a multi-modal probabilistic prediction for urban traffic data to capture the different possibilities of future traffic states.

Type of Project: To be discussed with the student
Supervisor: Marko Maljkovic ([email protected])
Student: Open for application

In this work we aim to explore the feasibility of modeling the temporal evolution of congestion in a transportation network using a Gaussian Process (GP), with the goal of formulating a dual control problem tailored to achieve patrolling behavior for a fleet of drones. Leveraging historical data to establish a prior belief about the temporal distribution of congestion throughout the day, we seek to evaluate whether a GP-based model can effectively capture the inter-daily variability of congestion and support the design of adaptive patrolling algorithms grounded in dual control principles.

A good knowledge of Python is required, along with willingness to learn about Gaussian Processes and Dual Control.

Reference:
[1] https://arxiv.org/abs/2012.06276

Type of Project: To be discussed with the student
Supervisor: Marko Maljkovic ([email protected])
Student: Open for application

In this project we aim to investigate the possibility of using deep Q-Learning with recurrent networks to train a fleet of patrolling agents whose goal is to monitor congestion in a city region. The motivation for this methodology stems from the reprorted success in solving various Atari games using the DQN approach [1]. Given that we can formulate a patrolling problem in a simplified 2D grid environment that resembles the standard setup of Atari games, we want to investigate if the proposed algorithms in the literature can help achieve good patrolling behaviour of the whole fleet.

A good knowledge of Python is required, along with willingness to learn about Reinforcement Learning

Reference:
[1] https://arxiv.org/pdf/1507.06527

Type of Project: Bachelor / Master Project
Supervisor: Ran Chen ([email protected])
Student: Pablo V. Fernandez; Georgio Sawaya

Ride-hailing services have experienced rapid growth due to their convenience, affordability, and technological advancements. However, maintaining fair competition and effective market regulation is essential for the long-term sustainability of the industry. To support market analysis, this project applies game theory to model a duopoly ride-hailing market, where two competing companies strategically distribute their fleets across multiple regions to maximize profits. We develop a simplified aggregate model and formulate the problem as a two-player convex game, aiming to analyze the existence and uniqueness of equilibrium and its implications for market dynamics.

The project involves theoretical analysis within a convex game framework and numerical implementation for model validation. Applicants should have experience implementing convex optimization algorithms in Python or MATLAB, and prior knowledge of game theory is beneficial for the derivation.

Type of Project: Semester Project
Supervisor: Weijiang Xiong ([email protected]) 
Student: Liam Gibbons

Forecasting the future traffic states is a fundamental task for the management of transportation systems. While many deep learning methods have been proposed to address this problem, most of them only focus on predicting most probable values for the traffic states. In reality, the evolution of traffic is highly dynamic, and the sensor measurements inevitably contains noises. As a result, a point estimation is often insufficient to represent the possible future traffic states and provide grounds for decision making in traffic management. The idea to overcome this problem is to predict an uncertainty score in addition to the expected value, with which we can know how much the predicted value may vary.

In this project, the student will learn to adapt existing uncertainty prediction methods into traffic forecasting problem, and implement both epistemic uncertainty and aleatoric uncertainty prediction (from [1]) for traffic forecasting models. The student will also learn to combine these uncertainties to compute confidence intervals, which will also be visualized and evaluated. Depending on the progress, the project can include more contents in neural network design (e.g., anchor-based regression), auxiliary learning tasks (e.g., missing value completion or imputation). To smoothly complete the project, the student is expected to have knowledge in deep learning and some hands-on experiences with python and deep learning libraries.

Reference: [1] What Uncertainties Do We Need in Bayesian Deep Learning for Computer Vision?. Alex Kendall, Yarin Gal. https://arxiv.org/abs/1703.04977

Type of Project: To be discussed with the student
Supervisor: Yura Tak ([email protected])
Student: open for application

UAV (Unmanned Aerial Vehicles) surged recently as a promising solution for aerial traffic monitoring. However, the sensors face the occlusions that can occur due to high buildings, trees, and other obstacles in the target area. Such occlusions decrease the performance of the vehicle detector. To achieve accurate monitoring with an occlusion-aware vehicle detector, we aim to segment the occlusions present on the road from the drone videos. To do so, the student will leverage state-of-the-art segmentation model, such as SAM [1], to segment the roads and the occlusions.

The student should have a good programming skills in Python. Previous experience with machine learning and computer vision is a plus.

[1] Segment Anything, Kirillov et al., https://arxiv.org/abs/2304.02643

Type of Project: To be discussed with the student
Supervisor: Lynn Fayed ([email protected])
Student: open for application

High-capacity on-demand station-based micro-transit provides a flexible and convenient service with generally lower waiting time and trip delays compared to fixed-schedule public transit. In areas where public transit systems are underperforming or their area coverage is deficient, on-demand micro-transit services have the potential to improve service quality and increase accessibility among travelers. This however requires rethinking the passenger-to-vehicle matching and the vehicle dispatching algorithm to ensure the synergy between on-demand and fixed-schedule services.

In this project, students will develop an optimization framework that improves the on-demand vehicle-to-passenger assignment by utilizing historical bus lines data. The goal is to ensure that on-demand services provide sufficient support for available fixed-schedule bus operation by: (i) prioritizing areas with low public transit coverage, (ii) improving serviceability, and decreasing passenger waiting time in stations where public transit service is deficient, and (iii) reducing trip delays by routing vehicles to areas with low congestion. For this project, students are expected to have some knowledge of Python.

Type of Project: To be discussed with the student
Supervisor: Georg Anagnostopoulos ([email protected])
Student: open for application

Micromobility is becoming increasingly prevalent in cities. As a consequence, microvehicles, such as motorcycles, scooters and bikes, often share the same road space with conventional cars. We will refer to this coupling of a predominantly lane-based host system with a layer of lane-free parasitic flows as multispecies traffic. There are good reasons to believe that multispecies traffic is macroscopically a porous flow, where microvehicles may bypass congestion by percolating through the car traffic exactly lilke water percolating a porous substrate. In other words, multispecies traffic is an example of directed percolation (DP). DP entails the existence of a nontrivial, far from equilibrium phase transition from a disordered subcritical phase to an ordered supercritical phase that depends on a temperature-like control parameter and is governed by a universal power law relation. We argue that the temperature in our model is connected in a subtle way with the noise in the Vicsek model of collective motion [1].

In this project the students will be exposed to simulation techniques, as well as to more advanced concepts from complexity science, such as universality, criticality, ordering and (nonequilibrium) phase-transitions. Our focus is to address the research question of how, as we approach the so-called thermodynamic limit, the apparently chaotic dynamics of multispecies traffic comply to an underlying order, which can be precisely captured by a single, universal number known as the critical exponent. Critical exponents matter because they are surpisingly invariant to different system configurations and are also independent of the specific problem formulations.

References: [1] T Vicsek, A Czirók, E Ben-Jacob, I Cohen, O Shochet, Novel type of phase transition in a system of self-driven particles. Phys. Rev. Lett. 75, 1226–1229 (1995).

Type of Project: To be discussed with the student
Supervisor: Marko Maljkovic ([email protected])
Student: open for application

In this project, we focus on the concept of dual control for exploration and exploatation (DCEE), utilized for traffic monitoring in a city region with a time-varying spatial distribution of congestion. Due to their omnipresence and overall flexibility, we envision using a fleet of drones across the region that can jointly observe and share the local information about the state of the city, and aim to design an optimal search framework for localizing the congestion epicenters. As the locations of the epicenters are unknown, we have to rely on Bayesian inference to estimate them.

The goal of this project is to design and test a multiagent, probabilistic, optimal search framework originally proposed by [1] in a simplified simulated setup comprising a square-like city region and multiple agents that can move across the region. Students are expected to have good knowledge of Python and willingness to learn the basics of Model Predictive Control (MPC) and related optimization libraries.

References: [1] https://arxiv.org/abs/2012.06276

Type of Project: To be discussed with the student
Supervisor: Pengbo ZHU ([email protected])
Student: open for application

Autonomous Mobility-on-Demand (AMoD) systems are revolutionizing urban transportation by offering timely, door-to-door service and alleviating traffic congestion. Efficiently managing the distribution of empty vehicles, known as vehicle rebalancing, has emerged as a critical operation to ensure seamless service and customer satisfaction.

In this project, you will have the opportunity to dive into the world of AMoD systems and contribute to solving the empty vehicle rebalancing challenge. Leveraging the concept of potential fields, you will develop an algorithm that optimizes the distribution of empty vehicles. The potential field approach involves designing attractive potentials that pull vehicles towards high-demand regions and repulsive potentials to prevent oversupply by avoiding vehicle clustering. We welcome motivated students with control or robotics background, proficient coding skills, preferably in Matlab.

Type of Project: To be discussed with the student
Supervisor: Minru Wang ([email protected])
Student: open for application

Drones can be used to deliver packages in an urban environment. To overcome the coverage limitation imposed by their battery life, drones may ride on buses for a segment of their journey to conserve energy. The objective of this project is to develop an optimization framework where drones ride on buses with predefined stops and schedules in order to minimize the total flying distance, and compare the service level with a scenario where drones do not ride on buses.

Experience with Python is necessary. Knowledge on classical Vehicle Routing Problems and related problems is an asset, but not mandatory. The student(s) will get a chance to learn about Pick-Up and Delivery Problems and related optimization tools in Python.

Type of Project: Bachelor Project
Supervisor: Gustav Nilsson ([email protected])
Student: Fletcher Collis and Luca Liuzzi

In this project, you will implement and simulate ride-hailing operations (like Uber, Lyft) in an open-source micro simulator for traffic, SUMO https://eclipse.dev/sumo/. As a first part, to familiarize yourself with how to use the Python API for controlling SUMO, you will simulate the service in a toy example. For the second part, there is a fully calibrated scenario for the city of Turin without ride-hailing, https://github.com/marcorapelli/TuSTScenario . By assuming that a fraction of the trips are made through ride-hailing, different performance metrics of the ride-hailing service can be investigated. If time permits, further investigations can also be done, such as the benefits of pooling passengers.

For the project to be successful, you need to have decent general data science and scripting skills, e.g., you should be able on your own (and perhaps some googling) to write a Python script to plot a time series from data stored in an XML file reasonably quickly. Some Linux/Unix and Git experience is beneficial too. No previous experience with SUMO is required.

Type of Project: Bachelor Project
Supervisor: Marko Maljkovic ([email protected])
Student: Léo Wunderli, Oscar Goudet and Marius Pécaut

In this project, we aim to investigate how the task of monitoring traffic phenomena in a city region can be tackled using a swarm of drones. For a simplified case study describing an urban environment via graph, we are interested in designing a patrolling algorithm [1] that helps increase the heterogeneous coverage of the area while taking into account the monitoring capabilities of the drones.

We will begin by adapting the existing simulation environment that operates one drone and then proceed to design, test, and compare different coordination algorithms to control a fleet of drones. Students will have the opportunity to investigate different rule-based algorithms and, if interested in Deep Learning, to gain hands-on experience in designing some learning-based methods. Therefore, good knowledge of Python is required.

[1] Drone swarm patrolling with uneven coverage requirements. C. Piciarelli, G.L. Foresti, https://arxiv.org/pdf/2107.00362.pdf

Type of Project: Bachelor Project
Supervisor: Georg Anagnostopoulos ([email protected])
Student: Amélie Menoud and Orane Koenga

Disruptions of the traffic flow due to lane-changing events have long been considered by researchers as a significant determinant of congestion. From a modeling perspective, changing lane is often simplified as a discrete and instantaneous action, typically using discrete choice models. While valid for lane-based systems, such as highways, the assumption of perfect lane-discipline is too strong in multispecies urban traffic where cars share the roadspace with lighter and more flexible vehicles, such as motorcycles, bicycles, e-scooters, and other two-wheelers. Hence the research question is how we could implement lane-changing in multispieces urban traffic.

In this project, the students will work with an existing multispecies traffic simulator developed in our lab and try to expand its functionality by allowing the cars to perform lane-changes. Some of the concepts that we are going to focus on are self-organization, collision-avoidance, anticipation and stability. Any previous exposure to the principles of simulation is highly appreciated, but not necessary. Very good knowledge of Python is required.

Type of Project: To be discussed with the student
Supervisor: Zhenyu Yang ([email protected])
Student: open for application

The use of curbsides, typically avaiable for both buses and ride-hailing vehicles, plays a significant role in urban transportation management. The temporary parking of ride-hailing vehicles during pickups and drop-offs often lead to bus delays. Understanding this dynamic is vital for effective curb management and promoting bus services. Presently, there’s a gap in detailed knowledge on this topic. The student is tasked with utilizing microscopic traffic simulation software, such as Vissim, SUMO, or Aimsun, to model the interplay between bus and ride-hailing vehicles. There’s also the option to approach this study analytically, based on our assumptions regarding the road network and drivers’ behaviors.

A basic understanding of car-following models and traffic flow theory is beneficial, though not essential. A preliminary knowledge of Python or another programming language is highly recommended for this project. All results obtained from this study would be reproducible.

Type of Project: To be discussed with the student
Supervisor: Minru Wang ([email protected])
Student: open for application

With the introduction of ride-sourcing services in urban areas, this new market has notable influence on transit ridership and travel costs. While there is growing research on how these travel modes can coexist, we are very interested in a specific case where some spatial restriction on ride-sourcing vehicle access is in place. As part of this project, the student will model the operation of ride-pooling as a first-mile service to complement public transit service, possibly by extending our adapted implementation of [1] to include transit lines in a simplified network. We will then evaluate the system performance in a number of operational scenarios.

Good programming skills in Python and strong analytical skills are essential for this project. Experience in optimization is an asset.

[1] On-demand high-capacity ride-sharing via dynamic trip-vehicle assignment. Alonso-Mora et al. https://doi.org/10.1073/pnas.1611675114

Type of Project: Laboratory GC (MSc), 4 credits
Supervisor: Georg Anagnostopoulos ([email protected])
Student: Brian Alexis Salamin and Jordan Lucien Dessibourg

For many city dwellers, riding a two-wheeler (bike, e-bike, scooter, e-scooter, motorcycle) is an attractive alternative to driving a four-wheeler (car, e-car). Not only because of lower purchasing and/or operating cost, but also as a strategy to bypass congestion. Due to their smaller size and higher maneuverability, some of the faster two-wheelers, such as motorcycles, percolate forward by riding between the lanes. This phenomenon goes by various names, including “filtering”, “creeping”, “lane-splitting” or “virtual lane”, and its understanding requires a more generalized theoretical toolbox.

Lane-based traffic flow theory, as exemplified by car-following models, does not apply in the abscence of a clear following hierarchy and predetermined lanes. Inpired from research in pedestrian flow, we will investigate a hybrid model, where cars are treated as moving obstacles, and two-wheelers navigate between the four-wheelers by performing collision-avoidance. The students will have the opportunity to investigate and synthesize concepts, such as distance to collider, time to collision, and anticipation. Our objective is to simulate motorcycle dynamics in a hybrid environmment and to reproduce the formation of virtual lanes. Good programming skills are desirable.

Type of Project: Civil Systems (MSc), 4 credits
Supervisor: Marko Maljkovic ([email protected])
Student: Zachary Hansen and Lorenzo Ballinari

In this project, we aim to investigate how the task of traffic monitoring can be tackled using a swarm of drones. For a simplified case study describing an urban environment, we are interested in designing a patrolling algorithm that helps increase the heterogeneous coverage of the area while taking into account the monitoring capabilities of the drones. To make the algorithm map-invariant, we want to investigate if such a centralized agent operating the swarm can be trained via Reinforcement Learning [1].

Ideally, students should be familiar with (or show willingness to learn) Deep Learning and Reinforcement Learning concepts. Moreover, some hands-on experience with python and some deep learning libraries would be beneficial.

Reference: [1] Drone swarm patrolling with uneven coverage requirements. C. Piciarelli, G.L. Foresti, https://arxiv.org/pdf/2107.00362.pdf

Type of Project: Semester Project in minor in CSE (MSc), 8 credits
Supervisor: Yura Tak ([email protected])
Student: Thamin Maurer

UAV (Unmanned Aerial Vehicles) surged recently as a promising solution for aerial surveillance of multiple urban areas. However, the sensors face the occlusions that can occur due to high buildings, trees and other obstacles in the target area. Such partially occluded vehicle images decreases the performance of the vehicle detector. In order to achieve an accurate monitoring with an occlusion-aware vehicle detector, we aim at de-occluding the occluded vehicle images based on the drone videos.

The project will be articulated in two-steps. The first part consists of segmenting the target occlusions area of the vehicle images to de-occlude. The second part will focus on the generation of de-occluded vehicle images from the occluded vehicle images, exploiting GAN models.

The student should have programming experience in Python. Previous experience with machine learning and computer vision is a plus.

Type of Project: Civil Systems (MSc), 4 credits; Construction project (MSc), 4 credits
Supervisor: Minru Wang ([email protected])
Student: Sanad JOUHARI and Mya JAMAL LAHJOUJI

Matching algorithms for dial-a-ride services can provide good quality solutions that aim to serve as many requests as possible with short waiting time and a small detour. However, existing studies often focus on tactical aspects and analyze performance metrics resulting from simulations. To complement the theoretical analysis, the semester project will consider uncertainties associated with an operational station-based microtransit service.

To assess how a theoretical matching simulation can adapt to real-time operation, we invite interested students to incorporate aspects such as dwell time uncertainty and parking space constraints, and investigate how the service platform can adapt to these operational uncertainties and constraints. Good Python programming skill is required.

Type of Project: Bachelor project, 6 credits; Laboratoire GC (Msc), 4 credits
Supervisor: Minru Wang ([email protected])
Student: Manon Bertola and Yasser Tahiri

In this semester project, the student will work with a simplified ride-sourcing network where spatial demand imbalance patterns can be observed, for example, high flow into the city centre during morning peak hours. The objective is to study a dynamic model of the system, and optimize the service level by controlling the proportion of pooled trips, either by optimizing for an entire study period, or through a receding horizon approach. Good programming knowledge and experience with Python is essential. The project can be adapted according to the student’s skills and interests; therefore, no prior knowledge in control is required.

Type of Project: Civil Systems (MSc), 4 credits
Supervisor: Pengbo Zhu ([email protected])
Student: Jonas Affentranger and Kanamori Yuki

Mobility-on-Demand system is an emerging service within urban scenarios which shows its potential to reduce congestion at the same time optimize service quality for customers. A critical operational challenge is the problem of imbalance between vehicle supply and customer demand. In this project, we will investigate a bi-level control structure for repositioning empty vehicles. It benefits from efficient coordination between the actions of upper-level controllers which operate the aggregated traffic components (e.g. how many empty vehicles should move from one subregion to another according to estimated future demands), at the same time the self-management of individual vehicles at lower-level which can give relatively precise position guidance.

The objective is to develop the ability to formulate problems, design different control schemes, verify the performances and present comparisons and conclusions of what we find. Please note that good Matlab programming skill is required and highly appreciated.

Type of Project: Laboratoire GC (MSc), 4 credits; Bachelor Project, 6 credits
Supervisor: Georg Anagnostopoulos ([email protected])
Student: Sanad Jouhari and Evangelia Gkola

For many city dwellers, riding a two-wheeler (bike, e-bike, scooter, e-scooter, motorcycle) is an attractive alternative to driving a four-wheeler (car, e-car). Not only because of lower purchasing and/or operating cost, but also as a strategy to bypass congestion. Due to their smaller size and higher maneuverability, some of the faster two-wheelers, such as motorcycles, percolate forward by riding between the lanes. This phenomenon goes by various names, including “filtering”, “creeping”, “lane-splitting” or “virtual lane”, and its understanding requires a more generalized theoretical toolbox.

Lane-based traffic flow theory, as exemplified by car-following models, does not apply in the abscence of a clear following hierarchy and predetermined lanes. Inpired from research in pedestrian flow, we will investigate a hybrid model, where cars are treated as moving obstacles, and two-wheelers navigate between the four-wheelers by performing collision-avoidance. The students will have the opportunity to investigate and synthesize concepts, such as distance to collider, time to collision, and anticipation. Our objective is to simulate motorcycle dynamics in a hybrid environmment and to reproduce the formation of virtual lanes. Good programming skills are desirable.

Type of Project: Bachelor Project; Laboratory GC (MSc), 4 credits
Supervisor: Lynn Fayed ([email protected])
Student: Milesi Riccardo and Anne-Valérie Preto

On-demand micro-transit is a transportation alternative sharing similarities with both ride-hailing/ride-splitting from one side and public transit from the other. The relatively high capacity feature of the micro-transit vehicles is inherited from the classical public transit service operation. Nevertheless, the dynamic and flexible route and schedule of the operating micro-transit vehicles accentuates the great resemblance this service has with ride-hailing. However, even if the service structure of public transit and ride-hailing is well-established in the literature, our understanding of the on-demand micro-transit market dynamics is still deficient.

The aim of this project is therefore to use a simulation-based approach to understand the main features of this type of service. We will mainly focus on assessing the intricate relationships between demand, service rate, micro-transit vehicle occupancies, passenger detour, and driver trip length per passenger. The ultimate goal is to determine under what market conditions these services are efficient, and what is the network structure in which they could be beneficial. For this project, a good Python programming level is required.

Type of Project: Bachelor Project
Supervisor: Yura Tak ([email protected])
Student: Ogay Xavier and Mahmoud Dokmak

UAV (Unmanned Aerial Vehicles) surged recently as a promising solution for aerial surveillance of multiple urban areas. However, the sensors face the occlusions that can occur due to high buildings, trees and other obstacles in the target area. Such partially occluded vehicle images decreases the performance of the vehicle detector. In order to achieve an accurate monitoring with an occlusion-aware vehicle detector, we aim at de-occluding the occluded vehicle images based on the drone videos.

The project will be articulated in two-steps. The first part consists of segmenting the target occlusions area of the vehicle images to de-occlude. The second part will focus on the generation of de-occluded vehicle images from the occluded vehicle images, exploiting GAN models.