Autonomous vehicles promise safer roads, but introducing them into cities filled with conventional cars could create new challenges.

A 2026 study published in Case Studies on Transport Policy suggests that the safety benefits of automation depend heavily on how many automated vehicles are present and how they interact with human drivers.

Using Athens, Greece, as a case study, researchers examined how collision risk changes as cities gradually move towards fully automated transport. Their findings suggest that the transition itself could be more complicated than the technology's long-term potential.

Testing The Future Of Urban Traffic

Researchers Maria Oikonomou and George Yannis investigated 15 progressive deployment scenarios using microscopic traffic simulations. These models reproduce individual vehicle movements, allowing scientists to examine interactions across a busy urban road network.

The study included conventional vehicles, partially automated cars operating at SAE Levels 2–3, fully automated vehicles at Levels 4–5 and autonomous cars programmed to behave more aggressively.

The researchers then applied XGBoost machine learning and SHAP, an explanatory technique that helps identify which variables contribute most strongly to the model's results.

Partial Automation Brings Unexpected Risks

One of the study's most important findings concerns partially automated vehicles. In mixed traffic, cars equipped with Level 2 or Level 3 automation were associated with slightly higher collision risk. These systems automate certain driving functions but still require varying degrees of human involvement.

Their interaction with conventional vehicles creates a complex environment in which different driving behaviours coexist.

Fully automated vehicles produced a different result. As their proportion increased, simulated collision risk declined, suggesting that wider deployment could eventually improve urban road safety.

However, the benefits were neither immediate nor uniform across the network.

Congestion Remains A Major Problem

Automation did not eliminate the influence of traditional road safety factors.

Traffic density, queue length and sections controlled by stop signs emerged as particularly important predictors of collision risk.

This means that replacing conventional vehicles with autonomous models will not necessarily solve the problems created by congested streets or challenging infrastructure.

The simulations also examined aggressively behaving autonomous vehicles. Their overall influence on collision risk was relatively small and became more noticeable mainly when automated vehicles represented a substantial proportion of traffic.

Why Cities Need Different Strategies

Oikonomou and Yannis emphasised that the safety effects of automation depend on traffic composition, infrastructure and differences in driving behaviour.

Their findings indicate that cities need to consider the characteristics of individual road networks rather than assuming autonomous vehicles will produce identical results everywhere.

Infrastructure improvements and carefully planned deployment could become particularly important during the transition, when conventional and automated vehicles must share increasingly busy streets.

What The Findings Mean

The research offers a detailed simulation-based assessment rather than evidence from real-world collision trials. Its results therefore depend on the driving behaviours and traffic conditions represented in the models.

Nevertheless, the findings highlight an important distinction between the eventual benefits of full automation and the challenges of introducing it gradually.

For urban transport planners, the central question is not simply how quickly autonomous vehicles can replace conventional cars. It is how roads, traffic management and different levels of automation can work together safely throughout that transition.