The IRTAD 2026 conference highlighted a major shift in road safety practice, driven by advances in AI, connected vehicle data, and international collaboration. From explainable machine learning models delivering significant gains in fatality reduction to critical insights on data bias and ethical governance, the field is moving beyond prediction toward transparent, proactive systems. At the same time, new evidence on under-reporting, inequality, and policy effectiveness is reshaping how countries design and evaluate road safety strategies globally.
The conference demonstrated the power of international knowledge sharing, with Agilysis methods and approaches being applied and adapted globally. Our collaboration with Wouter Van den Berghe resulted in compelling evidence that traditional Key Performance Indicators add only marginal explanatory power beyond simple vehicle kilometers of travel in predicting fatalities — a provocative finding that challenges conventional wisdom about safety performance monitoring.
The UK's influence on international practice was evident in several domains. DfT colleagues Matthew Tranter and Megan Turner presented both the methodological challenges of linking police and health data and the strategic framework of the UK's new national road safety strategy. The linking police and health data feasibility study revealed that only 72% of ambulance-attended road traffic collisions could be matched to STATS19 records, with particularly low linkage rates for pedal cyclists (53%) compared to motorcyclists (82%). This systematic under-reporting has profound implications for understanding the true burden of road trauma.
The new UK strategy, the first in over a decade, responds to the UK's decline from 3rd to 4th in European road safety rankings with ambitious 2035 targets: 65% reduction in KSIs and 70% reduction in child KSIs. Built on Safe System principles with evidence-based identification of priority groups and behaviors, it demonstrates how national strategies can embed international best practice while addressing domestic challenges.
The AI Revolution: From Novelty to Necessity
Artificial intelligence applications at IRTAD 2026 revealed a field moving beyond proof-of-concept toward operational deployment. However, this maturation brings new responsibilities around transparency, equity, and ethics.
The most compelling AI application came from Addis Ababa, where researchers combined Random Forest and XGBoost models with SHAP interpretability techniques and iRAP economic evaluation. Their explainable machine learning framework achieved up to four times more fatality and serious injury reductions and 65-point higher benefit-cost ratios compared to traditional exposure-based prioritization. This demonstrates how AI can enhance not just predictive accuracy but decision transparency and resource efficiency.
However, the IVORY network presentations provided essential counterbalances to AI enthusiasm. Oscar Oviedo-Trespalacios's examination of synthetic data for road safety raised critical concerns about how generative AI might systematically overlook vulnerable road users by reproducing car-centered patterns from training data. The comprehensive ethical analysis from TU Delft researchers identified five AI task categories — forecasting, correlation analysis, detection, group formation, and optimization — each raising distinct concerns around responsibility, explainability, autonomy, justice, and privacy.
These ethical considerations are not subsidiary challenges but core governance issues that directly affect regulatory design, public legitimacy, and safety performance. As AI becomes integral to road safety systems, the field must develop institutional arrangements that make value assumptions explicit, support meaningful human control, and facilitate ongoing reflexive learning.
Connected Vehicle Data: Promise and Pragmatism
The expansion of connected vehicle and floating car data applications showcased both tremendous potential and important limitations. Vianova's analysis of central London traffic demonstrated that harsh braking events could predict future collision locations with 39% accuracy using just three months of behavioral data — equivalent to 430 days of traditional collision analysis for the same predictive power.
However, the most valuable contribution was methodological transparency about data limitations. Multiple presentations acknowledged challenges around data representativeness, provider reliability, and temporal-spatial variability. The Athens telematics and video fusion study revealed strong correlations between data sources (ρ = 0.95) while highlighting systematic differences due to spatial positioning near intersections.
This honest assessment of both capabilities and constraints represents maturation in the field's approach to novel data sources. Rather than overselling emerging technologies, researchers are providing realistic guidance on appropriate applications and necessary cautions.
Policy Implementation: From Evidence to Action
Several presentations demonstrated how research translates into policy practice. The French analysis of local road safety policies provided quantitative evidence that governance strategies must align with territorial context — Global Management approaches work best in rural areas, Proactive strategies in metropolitan contexts. Only 26% of French counties currently implement their context-optimal strategy, suggesting full alignment could reduce fatal crash rates by 24%.
Wouter Van den Berghe's analysis of fairness in road safety policy revealed that liberty and relevance dominate public acceptance patterns more than equity or feasibility considerations. This finding has practical implications for policy communication and design — technical effectiveness alone is insufficient if policies lack perceived legitimacy.
The WHO's preparation for the Global Status Report on Road Safety 2027 represents the most comprehensive global assessment of progress toward Decade of Action targets. As countries approach the 2030 deadline for 50% fatality reduction, this mid-term evaluation will provide critical evidence on which approaches are succeeding and where additional support is needed.
Technical Foundations: Enabling Policy Applications
While less visible than policy applications, technical methodological advances provide essential foundations for reliable analysis. The multi-camera timestamp correction work from Belgium demonstrated how systematic quality assurance procedures can dramatically improve trajectory analysis — interpolated timestamps produced 42.5-frame trajectories versus 24.6 frames for uncorrected data. Such technical rigor may seem mundane, but it enables the automated conflict detection that supports proactive safety assessment.
The compositional diffusion image synthesis for road infrastructure assessment addresses the "long-tail" problem in computer vision — how to ensure AI systems can recognize infrastructure types that are rare in training data. By factorizing generation into country themes and infrastructure attributes, researchers can synthesize targeted combinations missing from real datasets, supporting more robust AI-assisted assessment across heterogeneous national contexts.
The Hidden Inequality Revelation
A recurring theme across presentations was the systematic revelation of how conventional approaches miss or disadvantage vulnerable groups. The RAC Foundation's linkage of detailed vehicle characteristics to crash data revealed that 38% of drivers from the most deprived areas drive cars over 10 years old versus 26% from the least deprived areas — a systematic inequality in access to protective technologies that compounds other disadvantages.
Transport for London's analysis of work-related road risk estimated that nearly half of people killed or seriously injured on London's roads may be harmed in work-related collisions — a finding that challenges assumptions about casualty distribution and policy priorities.
These examples demonstrate how sophisticated analysis can reveal patterns invisible to traditional approaches. The implications extend beyond academic interest to questions of equity, resource allocation, and policy effectiveness.
Looking Forward: Challenges and Opportunities
The IRTAD 2026 conference showcased a field in transition. The move toward proactive safety systems represents both tremendous opportunity and significant challenges. Data integration across multiple sources offers unprecedented analytical capabilities while raising complex questions about privacy, algorithmic bias, and institutional responsibility.
The methodological sophistication demonstrated across presentations provides confidence that the technical foundations for Vision Zero are solidifying. However, the honest acknowledgment of limitations and the serious engagement with ethical implications suggest a field maturing beyond technical optimism toward practical wisdom.
For organizations like Agilysis, the conference demonstrated both the value of methodological leadership and the importance of international collaboration. Our contributions spanned from fundamental research on data quality to practical tools for policy implementation, showcasing the breadth of expertise needed to support evidence-based road safety practice.
