Thanks to a novel cognitive encoding framework developed by a multidisciplinary research team from the School of Engineering at the Hong Kong University of Science and Technology (HKUST), which was published in the Proceedings of the National Academy of Sciences (PNAS), self-driving cars will soon be able to "think" like human drivers in challenging traffic conditions.
Professor Hai Yang (right), Chair Professor of the Department of Civil and Environmental Engineering at HKUST, and his PhD student Hongliang Lu (left) from the Intelligent Transportation Thrust at HKUST(GZ), draw inspiration from neuroscience, human cognitive processes, and ethics to enable self-driving cars to “think” like human drivers. Image Credit: Hong Kong University of Science and Technology
By lowering total traffic risk by 26.3% and minimizing possible injury to high-risk road users like cyclists and pedestrians by an astounding 51.7%, this invention significantly enhances the safety of autonomous vehicles (AVs). With their danger levels reduced by 8.3%, even the AVs themselves profited, opening the door for a new framework to further automate vehicle safety.
One major drawback of current AVs is that their decision-making algorithms are limited to pairwise risk evaluations, which means they are unable to take into account interactions between many road users holistically.
This is in contrast to a skilled driver who, for instance, can maneuver a junction by putting pedestrian safety first while marginally jeopardizing the safety of other cars. The driver can then turn their attention to other cars in the area after pedestrians have been deemed safe. Humans' capacity for risk management is referred to as “social sensitivity.”
A research team led by Prof. YANG Hai, Chair Professor of the Department of Civil and Environmental Engineering at HKUST, developed a human-plausible cognitive encoding method by drawing inspiration from ethics, human cognitive processes, and neuroscience to equip AVs with social sensitivity. With the help of this technology, autonomous vehicles can see, assess, and act like considerate human drivers.
Three cutting-edge features are integrated into this new system:
- Individual Risk Assessment: Assesses the danger that each user of the road, including nearby vehicles, cyclists, motorcyclists, and pedestrians, faces. This entails evaluating their speed, separations from one another, and consistency in conduct. A child strolling close to the road, for instance, would be seen as high risk
- Socially Weighted Risk Mapping: Gives decision-making an ethical dimension by putting the safety of those who are at risk first. In practice, this implies that even when technical regulations permit it to continue, the AV may yield to a pedestrian
- Behavioral Belief Encoding: forecasts the impact of the AV's activities on the traffic situation as a whole. It takes into account, for example, whether a rapid lane change might result in nearby vehicles braking abruptly or worsening congestion
The study team used 2,000 benchmark traffic situations to assess the safety performance of this cognitive encoding approach. The findings indicated that the framework decreased total traffic risk by 26.3%.
Interestingly, these safety enhancements were accompanied by increased operational effectiveness. AVs with our approach performed driving duties 13.9% quicker on average in the aforementioned simulations, proving that performance and ethical driving can coexist.
By emulating the human capacity for holistic risk processing and moral reasoning, we enable AVs to behave more responsibly in ethically ambiguous situations, such as congested intersections or near schools. Our framework is designed to be flexible and adaptable to meet different regulations and social norms. For example, while some countries prioritize protecting vulnerable road users, others place greater emphasis on traffic flow efficiency.
Hai Yang, Chair Professor, Department of Civil and Environmental Engineering, Hong Kong University of Science and Technology
Professor Yang added, “Additionally, legal interpretations of accident liability vary across jurisdictions. Our system can adjust weightings, enabling AVs to drive like locals and making global deployment more feasible.”
The University of Washington, Southeast University, Beijing Institute of Technology, Tsinghua University, Tongji University, and Hong Kong University of Science and Technology (Guangzhou) collaborated on this groundbreaking study.
The study team’s next step is to create an extensive data set that reflects various regional driving trends and societal expectations. To assist upcoming integration and testing initiatives, they are also in talks with prospective collaborators.
Journal Reference:
Lu, H., et al. (2025) Empowering safer socially sensitive autonomous vehicles using human-plausible cognitive encoding. Proceedings of the National Academy of Sciences. doi.org/10.1073/pnas.2401626122.