Quo-Vadis Multi-Agent Automotive Research? Insights from a Participatory Workshop and Questionnaire DOI: XXXXXXX.XXXXXXX Conference: 16th International Conference on Automotive User Interfaces and Interactive Vehicular Applications; September 22–25, 2024; Stanford, CA, USA ISBN: 978-1-4503-XXXX-X/18
Pavlo Bazilinskyy, Francesco Walker, Debargha Dey, Tram Thi Minh Tran, Hyungchai Park, Hyochang Kim, Hyunmin Kang, Patrick Ebel · 2025
Read Quo-Vadis Multi-Agent Automotive Research? Insights from a Participatory Workshop and Questionnaire DOI: XXXXXXX.XXXXXXX Conference: 16th International Conference on Automotive User Interfaces and Interactive Vehicular Applications; September 22–25, 2024; Stanford, CA, USA ISBN: 978-1-4503-XXXX-X/18 by Pavlo Bazilinskyy, Francesco Walker, Debargha Dey, Tram Thi Minh Tran, Hyungchai Park, Hyochang Kim, Hyunmin Kang, Patrick Ebel (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-03281.txt.
Introduction
Pavlo Bazilinskyy and seven co-authors report a 2025 workshop study on where multi-agent automotive research should go next. Roads now mix self-driving cars, hand-driven cars, and exposed users such as walkers and cyclists. Most past work, though, studies one road user at a time.
A hands-on workshop and a survey ran at a car interface conference. Small groups of experts took part in each. They weighed blocks and chances in real-world tests, simulations, and computer models. The worth of multi-agent work is widely seen, they found, but hard method and tech blocks slow it.
Catalog record
| CREATOR | Pavlo Bazilinskyy, Francesco Walker, Debargha Dey, Tram Thi Minh Tran, Hyungchai Park, Hyochang Kim, Hyunmin Kang, Patrick Ebel |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 25 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-03281
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Quo-Vadis Multi-Agent Automotive Research? Insights from a Participatory Workshop and Questionnaire DOI: XXXXXXX.XXXXXXX Conference: 16th International Conference on Automotive User Interfaces and Interactive Vehicular Applications; September 22–25, 2024; Stanford, CA, USA ISBN: 978-1-4503-XXXX-X/18/06 Conference: 17th International Conference on Automotive User Interfaces and Interactive Vehicular Applications; September 21–25, 2025; Brisbane, QLD, Australia 17th International Conference on Automotive User Interfaces and Interactive Vehicular Applications (AutomotiveUI Adjunct ’25), September 21–25, 2025, Brisbane, QLD, Australia DOI: 10.1145/3744335.3758500 ISBN: 979-8-4007-2014-7/2025/09 CCS: Human-centered computing Human computer interaction (HCI) CCS: Computing methodologies Modeling and simulation Pavlo Bazilinskyy, Francesco Walker, Debargha Dey, Tram Thi Minh Tran, Hyungchai Park, Hyochang Kim, Hyunmin Kang, Patrick Ebel Abstract . The transition to mixed-traffic environments that involve automated vehicles, manually operated vehicles, and vulnerable road users presents new challenges for human-centered automotive research. Despite this, most studies in the domain focus on single-agent interactions. This paper reports on a participatory workshop ( N = 15) and a questionnaire ( N = 19) conducted during the AutomotiveUI ’24 conference to explore the state of multi-agent automotive research. The participants discussed methodological challenges and opportunities in real-world settings, simulations, and computational modeling. Key findings reveal that while the value of multi-agent approaches is widely recognized, practical and technical barriers hinder their implementation. The study highlights the need for interdisciplinary methods, better tools, and simulation environments that support scalable, realistic, and ethically informed multi-agent research. 1. Introduction To operate safely, automated vehicles (AVs) must function as responsible social agents within complex mixed-traffic environments. These settings typically include a variety of agents including AVs, manually driven vehicles (MDVs), and vulnerable road users (VRUs). The goals and intentions of these agents often differ and can be conflicting. A key challenge in such environments is the variability in situation awareness (SA) between agents. SA refers to understanding what is happening in one’s environment ( Endsley 1995 ) . When multiple road users interact, SA extends beyond individual cognition and becomes a shared construct. According to the Distributed Situation Awareness theory ( Stanton et al. 2006 ) , traffic safety depends on shared and dynamically updated knowledge across all involved agents. Achieving this requires real-time multidirectional information exchange. Yet, much of the current research focuses on isolated user interactions, leaving the dynamics of multi-agent SA underexplored. Real-world traffic involves complex social interactions such as group dynamics ( Jiang et al. 2018 ) and cooperative behaviors — e.g., an AV warning a pedestrian about another approaching vehicle ( Ter Borg et al. 2019 ) . These scenarios present challenges to the scalability and clarity of communication ( Dey et al. 2020 ; Tran et al. 2023 ) . For example, in situations with multiple pedestrians, an eHMI displaying “Walk” may be ambiguous with respect to its intended recipients ( Dey et al. 2020 ) . Multi-agent settings further complicate joint decision-making under uncertainty, demanding intelligent systems capable of processing and interpreting inputs from multiple (human or non-human) agents. These systems must ensure coherence in behavior, such as avoiding contradictory signals. Computational models ( Lorenz et al. 2024 ) can contribute to this goal by capturing human perception, decision making ( Ebel et al. 2023 ) , and cognitive processes ( Oulasvirta et al. 2022 ) , allowing deeper insights into behavioral adaptations in dynamic traffic. Initial single-agent modeling efforts involving drivers and VRUs highlight the promise of such approaches and underscore the relevance of computational modeling in the advancement of multi-agent research ( Jokinen et al. 2025 ; Wang et al. 2024 ) . Despite some progress, existing studies often rely on overly simplistic scenarios ( Kooijman et al. 2019 ; Fridman et al. 2019 ) , which do not capture the complexity of real-world multi-agent traffic scenarios. However, some efforts address this problem. Bazilinskyy et al. 2020 developed an open-source simulator that allows large-scale multi-user interaction in shared virtual traffic scenarios. This setup has been used to study collaborative behaviors such as pedestrian-AV eye contact ( Mok et al. 2022 ) and eHMI-guided collision avoidance ( Bazilinskyy et al. 2022 ) . Several recent works similarly showcased linked / distributed simulators that promote the investigation of multi-agent interaction research and enable immersive real-time interaction between users, offering rich insights into joint behavior ( Feng et al. 2023 ; Tran et al. 2024 ) . From a cognitive psychology perspective, the influence of group dynamics on SA ( Capallera et al. 2023 ) , workload ( Lee et al. 2020 ) , and trust ( Walker et al. 2023 ) remains largely unexplored. Moore and Gugerty 2010 took an initial step by analyzing group conversations during Tesla Autopilot rides, finding that most groups (94%) experienced shared feelings of vulnerability and engaged in collaborative meaning-making to form shared mental models of AV behavior ( Momen et al. 2023 ) . However, this work lacked objective behavioral measures. We argue that a clear gap exists in the automotive interaction research literature concerning multi-agent traffic interaction, particularly from human factors and design perspectives. In this paper, we present initial findings from a participatory workshop that systematically explored the current state of multi-agent research and examined the key challenges and requirements for conducting rigorous studies in the context of automated mobility. 2. Participatory Workshop 2.1. Method Figure 1 . Participants in the workshop engaged in one of the group work discussions (left) and the discussion at the end of the workshop (right). Faces are blurred for anonymity. The participatory workshop was organized as part of an academic conference on automotive user interfaces on 22 September 2024 during the AutomotiveUI conference in Stanford, CA, USA. Fifteen attendees of the conference participated in the workshop. The workshop (see Fig. 1 ) included an introduction, a presentation on existing multi-agent automotive research, and two group work sessions separated by a break on the topics of A. “Multi-agent research in real-world settings”, B. “Multi-agent research in VR simulations” and C. “Multi-agent modeling research in automotive”, live demos of the coupled simulator ( Bazilinskyy et al. 2020 ) . During group sessions, participants used sticky notes to summarize the main points. The workshop ended with group presentations and a summary. The study was approved by the Ethics Review Board of Eindhoven University of Technology and the participants (both in the participatory workshop and questionnaire) gave their informed consent to use their data for research purposes. 2.2. Results 2.2.1. Multi-agent research in real-world settings This group explored both the potential and the challenges associated with conducting multi-agent research in naturalistic settings. First, through analyzing AV driving videos and structured scenario mapping exercises, participants identified various scenarios where multi-agent interactions could realistically occur (see Fig. A2 ). They highlighted that research on communication in mixed-traffic scenarios needs to move beyond hypothetical interactions and should address genuine human emotions and behaviors, such as hesitation, anxiety, and trust, which significantly influence interactions in real-world contexts. The discussion highlighted the importance of social and traffic norms in shaping human behavior in daily traffic . For instance, pedestrians often decide to cross or hesitate based on the subtle movements of vehicles. Similarly, drivers engage in informal negotiations about the right-of-way at intersections. These examples suggest that human road users are highly dependent on shared, often