Socially sensitive self-driving cars reduce the threat to road users

Self-driving cars trained to react to hazards more like humans will cause fewer injuries in traffic accidents , according to a study that shows how driverless vehicles could be made safer.

Vulnerable groups such as cyclists, pedestrians, and motorcyclists saw their protection increased the most when self-driving cars used "social sensitivity" to assess the collective impact of multiple hazards.

The study, published in the US scientific journal Proceedings of the National Academy of Sciences , highlights the growing efforts to balance the efficient operation of autonomous vehicles with the need to minimize damage in the event of a collision.

The study comes as leading tech companies like Tesla, Google's Waymo, and Amazon's Zoox strive to deploy autonomous vehicles worldwide. These vehicles use a host of sensors and automated software to drive without human intervention. Manufacturers must train the vehicles to respond instantly to real-world dilemmas, such as what to collide with if a crash is unavoidable.

The question of the ethics of autonomous vehicles is gaining increasing interest, as their increasing use offers the potential to eliminate driver problems such as space miscalculations and fatigue.

The study suggests that behavioral methods could "provide an effective framework for autonomous vehicles to address future ethical challenges," according to its Chinese and American authors, led by Hongliang Lu of the Hong Kong University of Science and Technology.

"By building on social concern and human-like cognitive coding, we enable autonomous vehicles to demonstrate social sensitivity in ethical decision-making ," they state. "This social sensitivity can help autonomous vehicles better integrate into existing driving communities."

"Social sensitivity" involves being attuned, like human drivers, to the vulnerability of certain road users and being able to judge who is most likely to be seriously injured in an accident.

The researchers drew on data from neuroscience and behavioral science showing that humans navigate using a "cognitive map" to interpret and adapt to the world.

The scientists based their instructions for autonomous vehicles on a concept known as "successor representation," which encodes predictions about how different elements of an environment will interrelate over time and space.

They examined the results of applying their model to EthicalPlanner, a system used by autonomous vehicles to make decisions based on various risk considerations. The researchers modeled 2,000 benchmark scenarios, measuring the total risk of each by assessing the probability of a collision and the likely severity of harm to the people involved.

The scientists found that using their human-inspired model with EthicalPlanner reduced overall risks for all parties by 26.3% and for vulnerable road users by 22.9%, compared to using EthicalPlanner alone.

In collision situations, all road users suffered 17.6% less damage, a percentage that increased to 51.7% for vulnerable road users. Occupants of autonomous vehicles also fared better, suffering 8.3% less damage.

An independent group of experts consulted by the European Commission has called for autonomous vehicles to be programmed to ensure "fair risk sharing and the protection of basic rights, including those of vulnerable users."

According to Daniela Rus, director of the Computer Science and Artificial Intelligence Laboratory at the Massachusetts Institute of Technology, the latest study addresses the crucial question of "how to model safe and socially sensitive behavior in autonomous vehicles."

"The proposed framework... offers a potential path toward autonomous vehicles capable of handling complex multi-actor situations, while being aware of the varying levels of vulnerability of road users," Rus adds.

© The Financial Times Limited [2025]. All rights reserved. FT and Financial Times are registered trademarks of Financial Times Limited. Redistribution, copying, or modification is prohibited. EXPANSIÓN is solely responsible for this translation, and Financial Times Limited is not responsible for its accuracy.