The 4th Workshop on Automated Mobility in Mixed Traffic

Motivation and Aim: 

The integration of automated vehicles (AVs) into existing transportation systems is transforming mobility and introducing unprecedented challenges and opportunities in mixed-traffic environments. In such settings, AVs with varying automation levels coexist with human-driven vehicles (HDVs) and vulnerable road users (e.g., cyclists and pedestrians). While AVs rely on algorithmic perception, planning, and control, human road users act based on experience, reasoning, intuition, risk perception, and social cues. These fundamentally different decision-making mechanisms give rise to complex interaction patterns shaped by human behavior, expectations, and adaptation processes.

As mixed-traffic systems evolve, it becomes essential to understand not only the technical performance of AI-driven systems but also how they interact with, influence, and adapt to human behavior. These dynamic interactions may significantly affect traffic safety, efficiency, energy consumption, and societal acceptance, yet remain difficult to analyze and predict.

Data-driven, empirical, model-based, and simulation-based approaches empowered by advanced AI tools are critical for studying these dynamics. However, challenges persist, including limited model generalization, discrepancies between simulation and real-world behavior, scarcity of high-quality mixed-traffic datasets, and insufficient integration between AI system design and behavioral insights.

By bringing together interdisciplinary expertise, this workshop aims to share state-of-the-art findings, foster in-depth discussion, and explore solutions to advance mixed-traffic research toward robust, innovative, and societally responsive mobility systems.

Building upon the success and experience of previous versions of the workshops at ITSC 2025, ITSC 2024, and ITSC 2023, this fourth edition further pushes forward the research for automated mobility in mixed traffic by: 

• Providing a unique opportunity for knowledge sharing by gathering together notable researchers in the domain and experts from the leading data collection and vehicle automation companies;

• Showcasing the available emerging datasets, their formats, and structure, and discussing their limitations, and challenges for the current research;

• Showcasing and validating state-of-the-art modelling methods and assumptions, with mixed traffic flow datasets; 

• Identifying current research gaps and future research directions, as well as the opportunities for creating synergy between data-driven and theory-driven research;

• Presenting the new IEEE ITSS Technical Committee with its community website for sharing relevant resources (open-sourced datasets, simulation tools and platforms, and pertinent publications).

Participants of this workshop will have the opportunity to communicate with other researchers and experts face-to-face. The goals are to share best practices, discuss common problems that have not been addressed, and gain insights on future research directions, so as to stay ahead of the curve. Additionally, a set of relevant research resources, e.g., open-sourced datasets with detailed summaries, simulation platforms and tools, relevant publication list, will be shared with the participants after the workshop.

Topics of Interest: 

Interested researchers are invited to present their works on the following relevant topics, including but not limited to:

1) Data & Simulation:

·   Mixed traffic datasets, open data practices and open-source platforms.

·   Empirical studies, field tests, and simulation models.

·   Mixed traffic state prediction and system modeling (long/medium/short term).

·   Behavioral and interaction modeling in mixed traffic.

2) Human Interaction:

·   Human factors, trust and user perception of automated systems.

·       Human-AI co-adaptation and meaningful human involvement in mixed traffic.

·       Ethical considerations in mixed autonomy systems

3) Advanced AI in Automated Driving:

·       AI methods in mixed traffic research.

·       LLMs and VLMs applied to automated driving.

·       Sensing, perception, planning and control in automated driving.

4) Safety & Robustness:

·       Model robustness and generalization.

·       Safety assurance and uncertainty quantification.

·       Explainable and trustworthy AI in safety-critical environments.

5) System Impacts:

·       Impact evaluation methods of mixed traffic.

·       Empirical evaluation across automation levels.

·       Traffic flow safety, efficiency, and energy performance in mixed traffic.

·       Societal and policy implications of mixed autonomy. 

Speakers’ Topic and Bio

• Topic: How Much Can AV Traffic Smoothing Reduce Emissions?

Daniel Work, Chancellor Faculty Fellow, Professor in Civil and Environmental Engineering, Computer Science, Vanderbilt University, USA

• Abstract: 

Automated vehicle (AV) control has the potential to smooth traffic waves and reduce emissions, but field experiments at the scale of months or longer remain out of reach. Here, we ask: how much could an AV reduce emissions by smoothing its trajectory in real traffic? Using massive I-24 MOTION trajectory data, we construct a counterfactual benchmark that replaces each empirical trajectory with a smooth, feasible AV trajectory, constrained to remain within a maximum gap of its leader. With a 0.1-mile maximum gap, passenger cars show 9.8–14.1% reductions in CO₂ and substantially larger reductions in CO, HC, and NOx across lanes. Trucks show roughly twice the potential of cars, highlighting an attractive target for mixed-autonomy control.

• • Speaker’s Bio:

Dr. Dan Work is a Chancellor Faculty Fellow and Professor in Civil and Environmental Engineering, Computer Science at Vanderbilt University. Dr. Work pioneered methods for monitoring and controlling road traffic using vehicles, rather than fixed infrastructure, to sense and control road congestion. He experimentally demonstrated that "phantom" traffic jams can be eliminated via control of a small fraction of AVs in the flow.

• From the Lab to the Street – Open Source Software for Interactive Urban Scenario Driving

Mattia Piccinini,  Humboldt Post-doctoral Fellow at the Autonomous Vehicle Systems (AVS) Lab, Technical University of Munich (TUM), Germany

• Abstract: 

In this talk we will present an open-source autonomous driving software stack designed to enable interactive urban scenario driving, bridging the gap between controlled laboratory research and real-world deployment. Built upon Autoware, the system integrates perception, localization, planning, and control modules into a modular and extensible architecture suitable for rapid experimentation and validation. The approach is demonstrated on the TUM EDGAR vehicle platform, which serves as a full-scale testbed for evaluating interactive behaviors in complex urban environments. Experimental results highlight the robustness, flexibility, and reproducibility of open-source software in advancing safe and scalable autonomous urban driving.

• • Speaker’s Bio:

Mattia Piccinini is a Humboldt Postdoctoral Fellow at the Autonomous Vehicle Systems (AVS) lab at the Technical University of Munich (TUM). He received an M.Sc. in mechatronics engineering (cum laude) and a Ph.D. (cum laude) in autonomous systems from the University of Trento, Italy, in 2019 and 2024, respectively. He obtained the TUM Global Postdoctoral Fellowship (2024), the ITSS Best Dissertation Award (2025), and the Humboldt Postdoctoral Fellowship (2025). He was a visiting researcher at the Universität der Bundeswehr Munich. His research focuses on physics-informed motion generation and control of vehicles and mobile ground robots.

• Reinforcement Learning and Control Barrier Functions for Connected and Automated Vehicles

Maria Laura Delle Monache, Assistant Professor, Department of Civil and Environmental Engineering and the Institute of Transportation Studies, UC Berkeley, USA

• Abstract: 

In this talk, we will present a framework to estimate collision risk in mixed autonomy scenarios when the road is shared between human driven vehicles and autonomous vehicles. The talk will introduce a probabilistic risk assessment framework that uses spatiotemporal occupancy heatmaps and captures uncertainty in predicted interactions.

• • Speaker’s Bio:

Dr. Maria Laura Delle Monache is an Assistant Professor in the Department of Civil and Environmental Engineering and the Institute of Transportation Studies at UC Berkeley. Maria Laura Delle Monache is a member of the Standing Committee on Traffic Flow Theory and Characteristics of the Transportation Research Board. Dr. Delle Monache’s research expertise lies at the intersection of transportation engineering, mathematics, and control theory.

• Multi-objective control of Automated Vehicles considering as a necessity, not an option

Simeon Calvert, Associate professor, Smart & Automated Driving in the department of Transport & Planning, TU Delft, the Netherlands

• Abstract: 

Automated vehicles are often optimised to be safe and comfortable, as well as to obey regulations. Safety, comfort, efficiency, traffic flow, energy use, and compliance are all important, but they are not enough if the vehicle’s behaviour is no longer understandable, influenceable, or accountable to the humans affected by it. This presentation argues that multi-objective control of automated vehicles should start from the question of control: that what we as humans (as drivers, passengers and society) state, desire and expect of automated vehicles should be at the heart of development. Moreover, when optimised as such, all the other mentioned metric are also automatically achieved and optimally balanced.  Meaningful Human Control is introduced as a practical way to achieve this challenge. The central claim is that human-aligned control is not an additional objective to be traded off against performance; it is the condition that makes other performance objectives meaningful.

• • Speaker’s Bio:

Dr. Simeon Calvert is associate professor of Smart & Automated Driving in the department of Transport & Planning at the TU Delft. He is director and founder of the Automated Driving & Simulation (ADaS) research lab and co-director of the CityAI-lab for research on urban behaviour using AI. He is also a board member of the Centre for Meaningful Human Control over AI, the interdisciplinary research program for responsible autonomous technology. His research is focussed on the impacts of technology on road traffic through experimentation, conceptualization and simulation.

• Evolving Traffic Signal Control for Mixed-Traffic Deployment: From Vehicle-Centric DRL to Cooperative, Explainable and Human-Centric AI

Shen Wang, Assistant Professor, School of Computer Science, University College Dublin, Ireland

• Abstract: 

Deep reinforcement learning (DRL) has demonstrated substantial potential for adaptive traffic signal control, progressing from conventional rule-based systems such as SCATS and SCOOT to learning-based approaches such as IntelliLight and PressLight. However, much of this research remains predominantly vehicle-centric, whereas real-world mixed traffic requires traffic control systems to accommodate connected and autonomous vehicles (CAVs), human-driven vehicles, pedestrians and cyclists, while addressing efficiency, safety, fairness, communication overhead and trustworthiness.

This talk presents our recent research exploring how traffic signal control can evolve toward such mixed-traffic deployment. First, CoTV extends traffic signal optimisation to cooperative control, jointly coordinating traffic signals and CAV speeds in mixed-autonomy traffic to improve mobility, fuel efficiency and safety while limiting the number of CAVs required for coordination. Second, xTL introduces XAI-guided semantic state design, using explanations to identify lightweight yet informative traffic features and thereby reducing communication overhead by over 90% without compromising control effectiveness. We then move beyond vehicle-centric optimisation toward vulnerable road users (VRUs). SafeCrossLight incorporates pedestrian-related safety information directly into DRL state and reward design, while FairCrossLight further addresses unequal service between vehicles and VRUs through adaptive fairness-aware learning, jointly improving efficiency, fairness and temporal stability. Finally, we explore an LLM-in-the-loop framework that interprets multimodal traffic and DRL feedback to provide structured recommendations for state and reward refinement, reducing reliance on manual trial-and-error and enabling more systematic safety–efficiency optimisation. Together, these studies illustrate a progression from optimising traffic signals for vehicles toward cooperative, human-centric and trustworthy AI-based traffic control for mixed mobility systems.

• • Speaker’s Bio:

Shen Wang (Member, IEEE) is currently an Assistant Professor at the School of Computer Science, University College Dublin, Ireland. He received an M.Eng. degree from Wuhan University, China, and a Ph.D. degree from Dublin City University, Ireland. Dr. Wang is a member of the IEEE and has been involved with several EU projects as a co-PI, WP, and Task leader in big trajectory data streaming for air traffic control and trustworthy AI for intelligent cybersecurity systems. Some key industry partners of his applied research are IBM Research Brazil, Boeing Research and Technology Europe, and Huawei Ireland Research Centre through various EU and state fundings such as Science Foundation Ireland (SFI). He is the recipient of the IEEE Intelligent Transportation Systems Society Young Professionals Travelling Fellowship 2022. His research interests include connected autonomous vehicles, explainable artificial intelligence, and security and privacy for mobile networks.

• Title: Taming noisy point clouds generation: how to balance physics and foundation models

Valentina Donzella, Professor, SEMS Research Excellence Framework Research Culture Lead, AESIN Sensor Lead, School of Engineering and Materials Science, Queen Mary University of London

• Abstract: 

LiDAR is a key sensing technology to enable safe and robust automated mobility in mixed traffic. However, LiDAR data degrades remarkably due to adverse environmental conditions, among which fog is a critical and common one. In order to enable safe LiDAR based perception even under foggy conditions, there is the need of collecting a huge amount of LIDAR data in these conditions, however it is difficult to achieve enough controllability and variability in real-world datasets. In this context, high fidelity simulation of point clouds, systematically degraded due to different fog intensities can play a key role to enablebio the design of robust perception. In this talk we will explore point cloud data augmentation, starting from validated physics based noise models, to more innovative techniques deploying a novel physics-grounded Vision Foundation Model (VFM) framework.

• • Speaker’s Bio:

Valentina Donzella is currently a Professor and the Head of the SPRING (Sensing and Perception for Intelligent Systems) Group, at Queen Mary University of London, U.K. Prof Donzella is the PI on several funded projects on automotive perception sensors, including the EU Horizon ROADVIEW.  She was previously awarded a Royal Academy of Engineering Industrial Fellowship on camera sensors.She is first author, coauthor, and last author of several journal articles on top-tier sensors and intelligent systems' journals. Her research interests include perception sensors for intelligent systems, machine learning, noise modelling, data quality, LiDAR, intelligent vehicles, integrated optical sensors, sensor fusion, and silicon photonics.

Prof. Donzella was the chair of the IEEE ITSS ICVES conference (2025), Associate Editor of the IEEE T-ITS, and is the founder and chair of the IEEE ITSS UK & Ireland chapter. She is a Full College Member of EPSRC and a Senior Fellow of the Higher Education Academy.

Resource Repository

The online resource repository for sharing relevant Datasets, Simulation Platforms, and Publications on Automated Mobility in Emerging Mixed Traffic can be accessed at https://qiqiqi.gitbook.io/mixed-traffic and https://github.com/IEEE-ITSS-OpenHub/Resource---Emerging-Mixed-Traffic-of-AV-and-HDV.

If you want to share relevant resources with the research community, please contact the workshop organizers.

Call for IEEE ITSS Peer Review Experience Survey Participation:

The IEEE ITSS Peer Review Experience Survey seeks feedback on current review systems for IEEE ITSS journals and conferences. The results will be used in aggregate to improve reviewer training, review-system guidance, reviewer matching, and reproducibility practices. Please do not include manuscript IDs, paper titles, reviewer names, editor names, or confidential information from unpublished manuscripts. Participation is voluntary. The survey does not ask for your name or email address.  Estimated completion time is 15-30 minutes for one role and may be longer for multiple roles. 

If you are willing to participate, please fill out the form at https://v.wjx.cn/vm/mPosirN.aspx.

Previous Editions 

At IEEE ITSC 2023, 2024 and 2025, the organizers hosted a previous edition of this workshop:  

https://www.itsc2025.mixedtraffic.org/

https://www.itsc2024.mixedtraffic.org/,

https://sites.google.com/view/itsc2023-mixed-traffic.

Acknowledgment

The workshop is supported by the interdisciplinary research community and IEEE ITSS Technical Committees of Automated Mobility in Mixed Traffic.