Event

Artificial Intelligence is rapidly transforming traffic safety research, development and decision-making.

From advanced driver support systems and automated driving to crash analysis, behavioural modelling and safety management, AI is opening entirely new possibilities. At the same time, it raises fundamental questions about trust, responsibility, data quality, regulation and human interaction.

This SAFER Research Day brings together leading researchers, industry experts and public-sector actors to explore not only what AI can do, but where it creates meaningful traffic safety value, where limitations remain, and what questions the community should focus on next.

Welcome to sign up here before September 23 COB!

Welcome!

PROGRAM

From 08:30 Registration, coffee and networking

09:00 – 09:05 Welcome and framing, Magnus Granström, SAFER

Setting the scene:

• Why AI and why now?
• What questions are emerging across the traffic safety community?
• What should we learn together?

SESSION 1

09:05 – 09:35 Keynote address: 

Understanding AI – and choosing the right approach: When Does AI Actually Create Value?
Christian Berger, University of Gothenburg

AI is increasingly applied to complex traffic safety challenges, but when is it actually the right tool? This keynote explores how to navigate between AI, machine learning and more traditional methods when developing safety-critical systems, and how to avoid adding complexity where it is not needed.

09:35 – 09:50 From Trustworthy AI Principles to AI Regulation: What Does It Mean for Traffic Safety?
Krishna Ronanki, Chalmers University of Technology

Many traffic safety applications, including ADAS and automated driving functions, are likely to be affected by the EU AI Act. This presentation introduces the principles behind trustworthy AI, explains the emerging regulatory landscape, and discusses what the transition from voluntary principles to legal requirements means for future traffic safety research and development.

09:50-10:05 From methods and data to meaningful predictions
Berenice Gudino, Chalmers Industriteknik

AI can support traffic safety research through advanced analysis, prediction and pattern recognition, but the value depends on choosing the right method for the right problem. This presentation brings a methodological perspective on applied AI, machine learning, deep learning and natural language processing, with a focus on how these tools can support traffic safety analysis and decision-making, and where their limitations need to be understood.

10:05 – 10:15 Can AI Read Accident Reports Better Than We Can?: Exploring the Potential of Large Language Models in STRADA
Erik Svanberg, Svanberg & Svanberg

STRADA is one of Swedens’s most important road safety databases, but much of its value remains locked in free-text descriptions, sketches and manually processed information. This presentation explores how Large Language Models (LLMs) could support accident reporting, data structuring and safety analysis, while also raising important questions around trust, validation, transparency and data quality. The presentation is based on a new SAFER pre-study exploring the opportunities and risks of introducing AI into future road safety data systems. 

10:15-10:45 Networking break 
 

SESSION 2

AI in Practice – Lessons from Current Research

10:45 – 11:45 AI Project Spotlights 

A series of short project pitches from across the SAFER community.

Each presenter will answer: 

• Which traffic safety challenge are you trying to solve?
• What role does AI play?
• What is the most important insight you would like to discuss during lunch?

Opportunity to continue the dialogue with the project members during the break!

11:45 – 13:00

Lunch & Poster Session

Each poster addresses:

• What AI method is used?
• Which traffic safety challenge does it address?
• What decision, action or impact does it enable?

Participants are encouraged to engage with the project teams and contribute perspectives from their own organisations and domains.

SESSION 3

13:00 – 13:30: Keynote address: Humans in AI-Enabled Traffic Systems: From Driver Assistance to AI Co-Pilots to Autonomy
Bryan Reimer, MIT AgeLab

Drawing on decades of human factors research at the intersection of technology and policy, this keynote explores the interaction between humans and AI-enabled systems and the implications for future traffic safety. While higher levels of autonomy may define the long-term future, the coming decades will likely be shaped by vehicles that support, guide, and cooperate with people. This makes the co-pilot model a critical bridge between today’s driver assistance systems and tomorrow’s automated mobility.

13:30 – 13:45 The Dangers of AI-Extrapolation of Normal Driving Data to Crashes 
Jonas Bärgman, Chalmers University of Technology

AI-based scenario generation and driver modelling can offer powerful tools for understanding traffic safety, but the value of these methods depends heavily on the data they are trained on. This presentation explores the risks of using normal driving data to extrapolate towards rare and safety-critical crash scenarios. It highlights key challenges around representativeness, model validity and the limits of data-driven approaches, while also reflecting on how AI methods, such as reinforcement learning combined with knowledge of driver cognition, may help bridge some of these gaps.

13:45 - 14:30 Community Reflection: What Should the Traffic Safety Community Focus on Next?

A facilitated dialogue bringing together keynote speakers, poster presenters and participants.

Discussion themes:

• Where does AI create the greatest traffic safety value?
• What are the most important unresolved challenges?
• What should we collaborate on as a community?
• What new research questions are emerging?
• What capabilities will be needed over the next five years?

14:30 End of event and networking break
 

Projects in the Poster Session

STIG - Smart Technology for Innovative Data Collection and Analysis for Pedestrian and Bicycle Paths
Daniel Rudmark, VTI

STIG explores how AI can turn infrastructure data into concrete safety improvements. By analysing images of pedestrian and cycling paths, the project helps municipalities detect gravel, cracks and potholes, prioritise maintenance and prevent accidents before they occur. The project highlights how AI can support proactive safety decisions beyond the vehicle itself.

Deep Multimodal Learning for Automotive Applications 
Selpi, Chalmers University of Technology

This project aims to create multimodal sensor fusion methods using AI methods for advanced and robust automotive perception systems. The project will focus on three key areas: (1) Develop multimodal fusion architectures and representations for both dynamic and static objects. (2) Investigate self-supervised learning techniques for the multimodal data in an automotive setting. (3) Improve the perception system’s ability to robustly handle rare events, objects, and road users.

Safety-Enhanced and Physics-Informed Foundation Models of Vehicle Behavior (SAFE-FM)
Yinan Yu, Chalmers University of Technology

SAFE-FM develops physics-informed AI foundation models to make automated vehicle validation more reliable and scalable. By combining AI learning with physical constraints, the project aims to improve prediction, interpretability and testing of safety-critical scenarios. It highlights how AI can support safer automated driving through better simulation, validation and proactive risk reduction.

Real and synthetic scenarios generated for the development, training, virtual testing and validation of CCAM systems (SYNERGIES)
Fredrik Warg, RISE

SYNERGIES focuses on reliable data and scenario generation for safe automated mobility. By developing open datasets, data-processing tools and a European scenario dataspace, the project supports training, virtual testing and validation of CCAM systems. It highlights how real and synthetic data can strengthen safety assurance for future automated vehicles.

VIASAFETY
John-Fredrik Grönvall, Chalmers Industriteknik

VIASAFETY uses AI methods to analyse data from electric micro vehicles such as e-scooters, cargo bikes, bicycles and e-mopeds. The project aims to identify risk zones and support better traffic and urban planning for low-speed mobility. It highlights how AI and connected vehicle data can improve safety for new forms of urban transport.

FATAL – Improved traffic safety at rural 2+1 intersections through data-driven insights and smart infrastructure
Ellen Grumert, VTI & Amritpal Singh, Viscando

The project team uses smart 3D sensor data to analyse behaviour and detect risky patterns at rural 2+1 road intersections. The project shows how AI-enabled infrastructure can support earlier risk identification and help traffic authorities take proactive safety measures where crash risks are often overlooked.

Intelligent, interactive and connected next generation real time driver assistance system (I2Connect)
Paul Hemeren, University of Skövde

The I2Connect project has developed the further driver connection between AI and HMI in relation to ADAS, driver monitoring (DMS), risk levels, and evidence theory. The relation to truck drivers also included three different drives to see eventual differences. Anticipatory real-time capabilities, including predictive processing, are vital for interactions with AI systems. This approach paves the way for more adaptive and personalized Human-Machine Interface (HMI) interactions, a significant departure from existing ADAS systems that typically offer static and non-personalized HMIs.

Beyond vision: A unified transformer with bidirectional attention for predicting driver perceived risk from multi-modal data
Kun Gao, Chalmers University of Technology

This work uses explainable AI to predict risky behaviour in automated vehicles and identify which road environment factors matter most. By combining machine learning with transparent decision-making, the study explores how AI can support safety assessment while maintaining trust, interpretability and actionable insights. 

AI-Based Video Analysis in Traffic Safety Research
Carl Johnsson, Lund University

Lund University brings long-standing experience of using AI-based camera and video analysis as a foundation for traffic safety research. Their contribution highlights how visual data can be transformed into research insights about behaviour, interactions and safety-critical situations, while also reflecting on methodological challenges such as data quality, interpretation, validation and what AI-based observations can, and cannot, tell us. 

Carl will also briefly introduce the new ROAD4SAFE project, focused on predicting and avoiding road crashes based on AI and big data.
 

ABOUT SAFER RESEARCH DAYS

SAFER Research Days aim to share insights, engage in meaningful discussions, present project results, identify next steps, strengthen networks, and gain new inspiration through thematic sessions, deep dives, guest speakers, and working group presentations.

Our Research Days are designed to achieve several important objectives:

• Disseminate knowledge: Share the valuable insights and discoveries generated from our diverse project portfolio.
• Engage in meaningful discussions: Participate in panel dialogues and workshops to delve into our partners’ findings, plan future steps, and emphasize practical applications in society as well as potential new projects.
• Identify next steps for projects: Explore potential next steps, paving the way for innovative proposals that drive progress and impact.
• Strengthen collaborative networks: Connect with peers, partners, and experts to exchange ideas and build stronger relationships for future cooperation.

In addition to these goals, our Research Days will feature:

• Thematic focus: Participate in sessions that focus on specific themes relevant to our research community.
• Deep dives and workshops: Engage in detailed explorations of specific topics to gain deeper understanding and insights.
• Guest speakers: Gain inspiration and knowledge from guest speakers who are leaders in their respective fields.
• Working Group presentations: Hear from various working groups about their ongoing projects and achievements.
• New inspiration: Discover new ideas and inspiration to drive your own research and projects forward.

These mini-conferences are for SAFER partners only!