Lectures

 
 
 
 
 
ETH AI Center & Institute for Transport Planning & Systems, ETH Zurich
Lecturer
September 2024 – Present Zurich, Switzerland

Computer Programming & Data Science – An Introduction with Python (101-0720-00L)

  • 2026 Spring Semester
  • 2025 Spring Semester

Microscopic Modelling and Simulation of Traffic Operations (101-0492-00L)

  • 2026 Autumn Semester
  • 2025 Autumn Semester
  • 2024 Autumn Semester

Projektübung Verkehr / Transportation Engineering Lab (103-0230-00 G)

  • 2026 Spring Semester
  • 2025 Spring Semester
 
 
 
 
 
Technische Universität Darmstadt
Tutor
October 2016 – February 2017 Darmstadt, Germany

Deterministische Signale und Systeme / Signal Processing (Prof. Dr. A. Klein, Prof. Dr.-Ing. M. Pesavento)

  • 2016 Autumn Semester

Students / Thesis

 
 
 
 
 
Semester Project, D-MATH, ETH Zurich
Cedric Zeiter
Semester Project, D-MATH, ETH Zurich
February 2026 – June 2026 Zurich, Switzerland

Topic: Who Breathes the Detour? A Coupled Traffic–Emission–Dispersion–Exposure Pipeline for an Alpine Transit Corridor

Abstract: Road traffic is among the largest sources of health-damaging air pollution in inhabited areas. Direct measurement of resident exposure at the scale of an entire corridor remains costly, and existing simulation approaches often target either high-fidelity local snapshots or regional averages, leaving the corridor scale underserved. We present a coupled, end-to-end pipeline that links four stages: microscopic traffic simulation in SUMO, per-vehicle emissions read from SUMO’s built-in HBEFA3 model, deterministic three-dimensional Eulerian diffusion solved with an implicit finite-difference scheme on a geometric vertical grid, and resident-level exposure expressed as an AQI over a synthetic population. The design deliberately trades the fidelity of full CFD dispersion for a workflow that runs end-to-end at valley scale on consumer hardware, and the complete implementation is released open source. We apply the pipeline to the Gotthard transit corridor in the Swiss canton of Uri, covering six villages, a high-traffic day and a lowtraffic day, and three pollutants (CO, NO2, PM2.5). Our findings indicate that PM2.5 dominates the AQI and remains localized within approximately 30m of the cantonal road, so that village layout, rather than population size, is the primary determinant of exposure. Reducing total traffic lowers the mean burden across a village but does not uniformly protect the most-exposed residents. Dynamic rerouting reduces congestion-related NO2 while simultaneously increasing CO through stop-and-go driving on side streets. The pipeline is transferable to other data-scarce transit corridors and is publicly available at https://github.com/cedriczeiter/FDM-SUMO.

Profile: https://www.linkedin.com/in/cedric-zeiter/

 
 
 
 
 
Master, Université Jean Monnet Saint-Etienne, Paris, France
Patrick Michael Barry
Master, Université Jean Monnet Saint-Etienne, Paris, France
February 2026 – July 2026 Zurich, Switzerland

Topic: A Replication Study of Safe Reinforcement Learning for Ramp Metering

Abstract: As urban transportation networks increasingly turn to reinforcement learning (RL) for dynamic ramp metering, the lack of methodological transparency poses significant safety and reproducibility challenges. This thesis conducts a comprehensive replication of a safe RL framework by Yang et al. (2025) using the SUMO microscopic traffic simulator, independently reconstructing key implementation details, including demand profiles and network geometry, that were not explicitly detailed in the original text. While the replication validates the framework’s core theoretical promise by demonstrating that its store-and-forward action replacement module successfully eliminates severe queue spillback and stabilises training, the original study’s absolute performance claims and travel time reductions could not be quantitatively reproduced. This divergence was primarily driven by the inability to determine the exact demand time series and structural network parameters used in the original study, forcing reliance on reverse-engineered approximations that meaningfully impacted the simulated traffic dynamics. Ultimately, this study confirms the functional viability of constrained RL for queue management but highlights that such advancements cannot be reliably verified or deployed without a communitywide shift toward open-source benchmarking and transparent reporting.

Profile: https://www.linkedin.com/in/patrick-barry29/

 
 
 
 
 
Master, University of Trento, Trento, Italy
Marco Prosperi
Master, University of Trento, Trento, Italy
February 2026 – July 2026 Zurich, Switzerland

Topic: [¬Re] Advanced Self Improving Ramp Metering Algorithm based on Multi-Agent Deep Reinforcement Learning

Abstract: This paper reports a replication attempt ([¬Re]) of Deng et al. (2019), who proposed a multi-agent deep reinforcement learning approach to ramp metering in SUMO. We provide a transparent methodological reconstruction of both the single-ramp and multi-ramp settings, comparing PPO and MAPPO with fixed-time and ALINEA baselines across three demand profiles. We were able to reproduce the main qualitative trends of the original study—the RL agent attains the highest return in two of the three demand scenarios, the cooperative multi-ramp agent restrains ramp queues, and the expected space–time congestion patterns are recovered—but the exact numerical results of Table I could not be reproduced: PPO speeds exceed the original by 16–25% and ramp queue lengths diverge substantially (∼145 m vs. 29–92 m). A systematic sensitivity analysis attributes these gaps to undocumented simulation parameters. We document the reconstruction process, the reproducibility barriers exposed, and release an open baseline for future benchmarking.

Profile: https://www.linkedin.com/in/marco-prosperi-05453515a/

 
 
 
 
 
Bachelor, D-INFK, ETH Zurich
Julian Isser
Bachelor, D-INFK, ETH Zurich
February 2026 – May 2026 Zurich, Switzerland

Topic: A Cyber-Physical System for Freeway Ramp Meter Signal Control Using Deep Reinforcement Learning in a Connected Environment: A Replication Study

Abstract: The goal of this replication study is to verify the claim by Hou et al. that, for freeway ramp metering signal control, the reinforcement learning algorithms Ape-X DQN, PPO and A3C outperform the baseline controllers no ramp metering, fixed-time control and ALINEA. The comparison is conducted under mixed near-congested/congested and extremely congested traffic conditions, using traffic throughput, average vehicle speed and per-vehicle averages for stops, fuel efficiency, CO2 emissions and NOx emissions as evaluation metrics.

Profile: https://www.linkedin.com/in/julian-isser/

 
 
 
 
 
Marc Gloor
Bachelor, D-BAUG, ETH Zurich
February 2026 – June 2026 Zurich, Switzerland

Topic: Assessing Ramp Metering for the A1 Corridor in the Canton of Aargau: A Simulation-Based Case Study

Abstract: Traffic congestion on the Swiss national road network continues to increase, particularly on heavily travelled corridors such as the A1 in the Canton of Aargau. As highway expansions face economic, environmental and political constraints, traffic management measures such as ramp metering offer a potential alternative for improving traffic flow. This thesis assesses the effects of ramp metering on a section of the A1 using microscopic traffic simulation in SUMO. A simulation model was developed based on detector measurements, vehicle-by-vehicle data and demand information from the cantonal transport model (KVM-AG). Two ALINEA controller parameterizations were evaluated under current traffic conditions and a forecast scenario for the year 2040. The results prove that ramp metering enhances corridor performance in both scenarios by reducing travel times, increasing mean speeds and lowering congestion levels. While the benefits under current demand conditions are modest, substantially larger improvements are achieved in the forecast scenario, where traffic demand exceeds highway capacity on several sections.

Profile: https://www.linkedin.com/in/marc-gloor-33809b216/

 
 
 
 
 
Raphael Benvenuti
Semester Project, D-BAUG, ETH Zurich
February 2026 – June 2026 Zurich, Switzerland

Topic: [RE] Ramp Metering for a Distant Downstream Bottleneck Using Reinforcement Learning with Value Function Approximation

Abstract: Scope of replication: This project replicates and evaluates the claims made by the original study of Zhou et al. (2020), Ramp Metering for a Distant Downstream Bottleneck Using Reinforcement Learning with Value Function Approximation. The original article proposes a reinforcement learning (RL) based controller for ramp metering and reports improved performance compared with conventional feedback-based ramp metering algorithms. Methodology: As no supplementary material, source code, or numerical demand data were provided, the full experimental setup was re-implemented from scratch in Python. The simulation environment was built using a Cell Transmission Model (CTM), while the RL agent was implemented and trained using PyTorch. Missing parameters were either assumed or reconstructed from the published figures and methodological description. Results: The no-control and reference PI-ALINEA scenarios could be reproduced closely. However, the reported performance advantage of the RL controller could not be confirmed when stronger benchmark controllers were introduced. In particular, optimized PI-ALINEA variants matched or outperformed the RL controller. What was easy: Implementing the basic CTM dynamics and reproducing the qualitative no-control and PI-ALINEA traffic patterns was comparatively straightforward once the main simulation parameters had been identified. What was difficult: The main difficulties were reconstructing undocumented model parameters, matching the original demand profile, implementing a stable RL training procedure, and comparing results against the original study, which reported mostly graphical outputs and no quantitative performance table. Communication with the original authors: No attempt was made to contact the original authors. No additional material, such as source code, numerical demand profiles, algorithmic parameters, or supplementary documentation, was available beyond the published article.

 
 
 
 
 
Master, Politecnico di Milano
Alessandro Bertelli
Master, Politecnico di Milano
January 2026 – June 2026

Topic: Studies of Reinforcement Learning Control in Intelligent Road Systems Physics-Informed Reward Shaping for Hybrid Model Predictive and Reinforcement Learning Ramp Metering

Abstract: Ramp metering, which regulates the flow of vehicles entering a freeway from its on-ramps, is one of the most effective tools for preventing freeway congestion. Deep reinforcement learning is a promising approach to this problem, but fewer than 2% of transportation simulation studies release their source code, so published claims can rarely be verified, and learningbased controllers offer no guarantee that operational limits, such as the maximum on-ramp queue, are respected. Model predictive control enforces such limits explicitly, but depends on the accuracy of its internal model and is computationally expensive. This thesis addresses both problems. First, a recently published hierarchical controller that combines model predictive control with deep reinforcement learning was reimplemented from its published description alone and released as open-source software. Second, on top of this verified implementation, a physics-informed variant was designed: during training, the METANET macroscopic traffic flow model predicts whether the proposed metering action will cause a queue or density violation within the next minute, and the agent is penalised before the violation occurs. The shaping signal is bounded relative to the environment reward, and training is stabilised by regularising the pre-activation outputs of the actor network. All controllers were evaluated on a benchmark freeway under two demand profiles and three noise levels, over ten stochastic seeds with paired statistical tests. The replication recovered the published behaviour, but only after numerous undocumented choices had been reconstructed, indicating that the reproducibility gap is primarily one of incomplete documentation. On the congested demand profile, the proposed controller reduced peak on-ramp queue overshoot to between one quarter and two fifths of that of standalone model predictive control, while keeping total time spent within 1.3% of the best baseline; its advantage over model predictive control was statistically significant in all six scenarios on queue violations and in five on total time spent. Since the prediction model runs only during training, this gain adds no computational cost at deployment relative to the original hybrid controller. The results suggest that constraint satisfaction and travel efficiency are largely compatible objectives, and that the evaluation protocol, including the treatment of random seeds and checkpoint selection, should be reported as part of any performance claim.

Profile: https://www.linkedin.com/in/alessandro-bertelli/

 
 
 
 
 
Master / Research Assistant, D-MATH, ETH Zurich
Jon Ruiz Martinez
Master / Research Assistant, D-MATH, ETH Zurich
October 2025 – March 2026 Zurich, Switzerland

Topic: [RE] Motorway Ramp-Metering Control with Queuing Consideration using Q-Learning

Abstract: Ramp metering is a problem from the domain of intelligent highway transportation systems, in which the inflow of vehicles onto highways from on‐ramps is regulated (metered) with traffic lights; if the highway is too congested, fewer vehicles can enter the vehicle [3, 4]. Ramp metering is shown to significantly reduce the congestion on highways, and various analytic approaches have been developed so far, including ALINEA [5], METALINE [6], SWARM [7], and HERO [8]. Reinforcement Learning (RL) [9] increasingly enjoys popularity in the control literature, as these numerical and learning‐based methods can achieve performances above conventional, analytic approaches [10, 11, 12, 13]. Therefore, a growing number of studies in the field of intelligent transportation systems control that try to address the ramp metering problem leverage the potential of RL [14, 15, 16, 17, 18]. At the same time, only a small quantity of these studies provide implementation details, code, data, and other material for reproducing these results, which lead to a reproducibility crisis in the field [19]. Therefore, we initiated a replication project, in which we aim to reproduce several works in the context of RL‐based control of intelligent transportation systems. In this work, we replicate the work ”Motorway Ramp‐Metering Control with Queuing Consideration using Q‐Learning” from Davarynejad et al. 2011 [1], in which three distinct reinforcement learning controllers are independently trained and evaluated for adaptive ramp metering. When replicating the study, we encountered several ambiguities and inconsistencies in the authors’ description, and cannot confirm the claimed performance gains of the multi‐agent approach, which according to our results are at most marginal and do not outperform the uncontrolled baseline scenario.

Profile: https://www.linkedin.com/in/jon-ruiz-mart%C3%ADnez-3402492aa/

 
 
 
 
 
Bachelor, Computer Science, ETH Zurich
Rasim Ibadov
Bachelor, Computer Science, ETH Zurich
October 2025 – February 2026 Zurich, Switzerland
Topic: Replication Study of a RL-based Traffic Optimization Paper Abstract: This study replicates and critically evaluates the claims made in the work by Deng et al. (2023), titled “Automated Traffic State Optimization in the Weaving Area of Urban Expressways by a Reinforcement Learning‐Based Cooperative Method of Channelization and Ramp Metering,” [1] published in the Journal of Advanced Transportation. The original study proposes a cooperative method that combines channelization and reinforcement‐learning‐based ramp metering and reports substantial improvements in lane‐wise vehicle speeds in a freeway weaving section. The primary objective of this replication is to verify whether the reported performance improvements can be reproduced under an independently implemented but methodologically faithful setup. To this end, we reconstructed the complete simulation environment and implemented all control strategies from scratch using the SUMO microscopic traffic simulation framework. Beyond a strict replication, this work also establishes a transparent and fully documented benchmarking environment, enabling systematic comparison with both classical and learning‐based controllers in future studies. Profile: https://www.linkedin.com/in/rasim-ibadov-b4a3b4235/
 
 
 
 
 
Master, Computer Science, ETH Zurich
Natcha Jengjirapas
Master, Computer Science, ETH Zurich
October 2025 – December 2025 Zurich, Switzerland

Topic: Reproduction Study Of RL Control for Intelligent Road Transportation Systems

Abstract: This semester project investigates the reproducibility of two reinforcement learning–based ramp metering approaches proposed by Lu et al. The first uses tabular Q-learning for local ramp metering to minimize total time spent (TTS) under on-ramp queue constraints, while the second extends this framework to coordinated ramp metering by incorporating an equity objective based on the variance of total waiting time (TWT). Both methods are independently reimplemented from the published descriptions using a macroscopic traffic simulation based on the asymmetric cell transmission model, without access to original code or supplementary data. Reproduced results are compared with those reported in the original studies using the same performance metrics and reconstructed demand scenarios. The results show that the baseline Q-learning controller largely reproduces the reported efficiency and queue management behavior, while the equity-aware extension captures the intended efficiency–equity trade-off with some numerical discrepancies. These findings support the main conclusions of the original studies while highlighting reproducibility challenges arising from incomplete methodological specification.

Profile: https://www.linkedin.com/in/njengjir/

 
 
 
 
 
Master, Applied Mathematics, ETH Zurich
Elyse Winstral
Master, Applied Mathematics, ETH Zurich
September 2025 – March 2026 Zurich, Switzerland

Topic: Individual, Urban Trip Reconstruction via BMS and Loop Detector Data

Abstract: Loop detectors and Bluetooth Media Access Control Scanners (BMS) have been used tomeasure various vehicle data in fixed locations and zones respectively. Recently, there has been interest in tracing individual vehicle trips with this data. While loop detectors are relatively common in European cities, BMS offers a more cost-effective way to gather additional information on traffic conditions. The aim of this thesis is to fuse BMS and loop detector data, and subsequently use the enriched data collection to reconstruct individual vehicle trips in a simulated, urban setting with a Markov Decision Process estimation method. Individual trip reconstruction may bolster traffic state analysis and prediction on meso- and macro-scopic levels.

Profile: https://www.linkedin.com/in/elyse-winstral/

 
 
 
 
 
Master, Electrical Engineering, ETH Zurich
Qiyuan Zhang
Master, Electrical Engineering, ETH Zurich
September 2025 – January 2026 Zurich, Switzerland
Topic: Reproduction Study Of RL Control for Intelligent Road Transportation Systems Abstract: This semester project investigates the reproducibility of two reinforcement learning–based ramp metering approaches proposed by Fares et al. and Wang et al.
 
 
 
 
 
Master, Mechanical Engineering, ETH Zurich
Omar Alami Badissi
Master, Mechanical Engineering, ETH Zurich
March 2025 – September 2025 Zurich, Switzerland
Topic: Coordinated Control Algorithm For Fairness In Ramp Metering Abstract: Traffic assistance schemes typically tend to focus on global system efficiency which can come at the cost of equity among users of the road. This paper proposes C-ALINEA, a new ramp metering algorithm that makes inter-ramp relationships explicit in order to tackle equity issues without losing throughput. As opposed to classical isolated controllers, C-ALINEA generates a coherent intelligence by connecting each ramp to two upstream and two downstream neighbors. The methodology employs SUMO microsimulations to evaluate the algorithm on the A10 ring road around Amsterdam, a heavily congested urban freeway with complex ramp interactions. C-ALINEA is compared against conventional ALINEA and other state-of-the-art approaches using calibrated traffic demand patterns from the typical A10 ring road. Performance metrics include mainline throughput, ramp waiting times, spatial equity indices, and system resilience. Results have shown that C-ALINEA can achieve a more fair sharing of congestion impacts over different entry points along the A10. With its five-ramp horizon, the algorithm has the possibility to predict downstream bottlenecks and harmonize the response on adjacent ramps. The results are encouraging in that coordination-based ramp metering appears to be a viable approach to socially acceptable traffic management solutions in urban settings. Profile: https://www.linkedin.com/in/omar-alami-badissi-97489920b/
 
 
 
 
 
Bachelor, Computational Science and Engineering, ETH Zurich
Cedric Zeiter
Bachelor, Computational Science and Engineering, ETH Zurich
March 2025 – May 2025 Zurich, Switzerland

Topic: Diffusion Modelling of Air Quality and Noise - Impacts from Detour Traffic on Uri Residents A Simulation-Based Case Study of Villages near the Gotthard Road Tunnel in Uri, Switzerland

Abstract: Traffic congestion near the northern entrance of the Gotthard Road Tunnel in the Swiss canton of Uri often leads to significant detours onto cantonal roads, diverting traffic through residential areas and raising concerns about the well-being of residents. While previous research and federally funded projects have largely focused on traffic flow optimisation, the perspective of affected residents has remained under-explored. This thesis addresses that gap by laying the foundation for a quantitative, resident-oriented assessment of the environmental externalities caused by traffic detours. Using traffic data generated by the SUMO simulation software, extended with a novel post-processing pipeline developed specifically for this work, the model estimates exposure levels of CO, NO2, PM2.5, and noise across six affected villages in the canton of Uri. The implementation is based on an explicit forward Euler scheme with central finite differences. Simulations were conducted over the course of a day for different dates, capturing both low- and high-congestion scenarios. The results show substantial variation in pollutant and noise exposure depending on the village and discuss what a reduction of traffic in Uri would change quantitatively.

Profile: https://www.linkedin.com/in/cedric-zeiter/

 
 
 
 
 
Master, Civil- & Traffic Engineering, ETH Zurich
Bertola Manon
Master, Civil- & Traffic Engineering, ETH Zurich
February 2025 – June 2025 Zurich, Switzerland

Topic: Impacts of Transit Detour Traffic on Local Accessibility

Abstract: This thesis investigates how congestion at the Gotthard tunnel’s north portal and resulting detour traffic affect local accessibility in the Urner Valley, and evaluates the effects of a policy restricting transit vehicles to the highway. A calibrated SUMO microsimulation model, based on loop-detector data, replicates traffic patterns for three representative 2024 case study days with varying congestion levels. Accessibility is assessed via simulated travel times to both the regional center (Altdorf) and the urban center (Zurich). Results show that road-based access degrades sharply on high-transit days, particularly in upper-valley towns, where public transport alternatives are weak. The policy experiment has minimal impact on low-demand days but significantly improves local accessibility on congested days, while increasing delays for through traffic. These findings highlight trade-offs between aggregate efficiency and local equity, and suggest the value of targeted, demand-responsive interventions.

Profile: https://www.linkedin.com/in/manon-bertola/

 
 
 
 
 
Bachelor, Mechanical Engineering, ETH Zurich
Justin Weiss
Bachelor, Mechanical Engineering, ETH Zurich
February 2025 – June 2025 Zurich, Switzerland

Topic: Infrastructural Control Algorithm for Fairness in Road Traffic Engineering

Abstract: This thesis investigates whether fairness and efficiency can be jointly achieved in urban traffic signal control at intersections. While efficiency has traditionally been the primary objective in signal control strategies, the consideration of fairness has received comparatively limited attention. To address this gap, a controller inspired by the adaptive systems SCOOT and SCATS, referred to as SCOSCA, is developed. SCOSCA incorporates the three core optimizers of these systems: an offset optimizer, a cycle length optimizer, and a green phase optimizer. Building upon this foundation, two fairness-augmented variants are introduced: SCOSCAFAIRV1, which integrates a fairness-aware green phase optimizer, and SCOSCAFAIRV2, which additionally includes a secondary real-time fairness controller. The controllers are evaluated on both a synthetic toy model and a realistic arterial road network in a case study, using the microscopic traffic simulation platform SUMO. Several fairness ideologies, Egalitarian, Rawlsian Harsanyian and Utilitarian, are formalized through respective performance metrics. Bayesian Optimization is applied to tune the controllers towards these fairness objectives. The findings show that fairness could be improved with minimal or no loss in efficiency; in the case study, efficiency could even be enhanced. Moreover, the results observed for SCOSCA, SCOSCAFAIRV1 and SCOSCAFAIRV2 under different optimization goals suggest that optimizing for a single performance metric often leads to improvements across multiple fairness indicators. This interdependence highlights the importance of explicitly incorporating fairness into the design of future traffic control algorithms.

Profile: https://www.linkedin.com/in/justin-weiss-36b779269/

 
 
 
 
 
Master, Computer Science, ETH Zurich
Lea Künstler
Master, Computer Science, ETH Zurich
December 2024 – June 2025 Zurich, Switzerland

Topic: Fair Perimeter Control using Queue Balancing

Abstract: This thesis investigates the integration of fairness into perimeter control strategies for urban traffic management. While perimeter control effectively reduces congestion by regulating inflow into protected zones, existing approaches typically prioritize system efficiency, often overlooking the equitable distribution of delays across users. A simulation-based evaluation demonstrates that basic perimeter control not only improves total and internal delays but also enhances fairness metrics, including the Gini coefficient and Jain’s fairness index. Building on this, queue balancing strategies inspired by computer network resource allocation were implemented to assess their impact on fairness. Results show that under uniform demand, queue balancing provides no consistent benefits. However, in varied demand scenarios, where certain entry points are more congested, fairness improvements were observed. A real-world scenario using the San Francisco network highlights the practical challenges of implementation, such as queue spillbacks and the need for more adaptive control mechanisms. Overall, the findings suggest that fairness-aware perimeter control is feasible and beneficial under realistic conditions, but further work is needed to ensure consistent and scalable deployment.

Profile: https://www.linkedin.com/in/leakuenstler465b59182/

 
 
 
 
 
Master, Industrial Engineering and Management Engineering, ETH Zurich & UPC Barcelona
Yangle Zhan
Master, Industrial Engineering and Management Engineering, ETH Zurich & UPC Barcelona
September 2024 – February 2025 Zurich, Switzerland

Topic: Fairness on Ramp Metering - Extending ALINEA for Equitable Access and Congestion Reduction

Abstract: Ramp metering systems effectively reduce freeway congestion but often face public opposition due to inequitable delay distribution among users. This study addresses this challenge by introducing EqALINEA, a fairness-enhanced extension of the ALINEA algorithm designed to balance efficiency and equity in freeway access. Using microsimulations in SUMO, we evaluate ALINEA, Upstream ALINEA, and EqALINEA across both a simplified network and a real-world case study of Barcelona’s Ronda de Dalt corridor. Results indicate that traditional ALINEA prioritizes mainline throughput at the expense of spatial equity, disproportionately delaying on-ramp users near bottlenecks. EqALINEA mitigates these disparities by incorporating fairness constraints, including maximum waiting thresholds and proactive queue management. This approach leads to a more balanced distribution of delays while maintaining overall traffic efficiency within the simulation time. The algorithm adapts effectively to urban environments such as the Ronda de Dalt, where infrastructure limitations and evolving mobility policies require equitable traffic management solutions. This study demonstrates the feasibility of integrating fairness into decentralized ramp metering strategies to improve public acceptance and align with sustainable urban planning objectives.

Profile: https://www.linkedin.com/in/yangle-zhan/

 
 
 
 
 
Master, Civil- & Traffic Engineering, ETH Zurich
Modest Jiang
Master, Civil- & Traffic Engineering, ETH Zurich
September 2024 – January 2025 Zurich, Switzerland

Topic: Esslingen‘s Traffic Light Upgrade: An Evaluation

Abstract: Seven traffic light signals received new control systems in the city of Esslingen am Neckar in Germany. This thesis investigates the performance and efficiency during peak hours of the old and new traffic light control system using SUMO simulations. The older traffic light control system is a signal control system that operates based on a predefined Signal Phase and Timing (SPaT) schedule but adapts its parameters in response to real-time inputs from traffic detectors. The new system is a signal control system that dynamically determines traffic signal phases using real-time detector data and an algorithm. The evaluation focuses on three critical metrics: travel time, waiting time, and delay. The data for the old system was provided in the form of a VISSIM simulation while the data for the new system was provided in form of detector data. The simulation runs on SUMO demonstrate that the newer traffic light signal control system outperforms the existing system across all evaluated metrics, indicating an overall improvement in traffic flow efficiency. These findings underscore the benefits of implementing the updated traffic light control system to enhance urban mobility and reduce congestion during peak traffic periods.

Profile: https://www.linkedin.com/in/modest-jiang/