In a groundbreaking move for road safety, Johns Hopkins University researchers have developed SafeTraffic Copilot, an AI-driven tool capable of predicting car crash risks and informing preventative measures.
Johns Hopkins University’s researchers have achieved a significant milestone in transportation safety with the development of SafeTraffic Copilot, an advanced artificial intelligence tool designed to predict and mitigate car crash risks across the United States.
The findings are published in the scientific journal Nature Communications.
“Car crashes in the U.S. continue to increase, despite decades of countermeasures, and these are complex events affected by numerous variables, like weather, traffic patterns and driver behavior,” senior author Hao (Frank) Yang, a professor of civil and systems engineering at Johns Hopkins, said in a news release. “With SafeTraffic Copilot, our goal is to simplify this complexity and provide infrastructure designers and policymakers with data-based insights to mitigate crashes.”
SafeTraffic Copilot leverages Large Language Models (LLMs), a type of AI proficient in processing, understanding and learning from extensive data.
The model was trained on diverse data sources, including textual descriptions of road conditions, numerical data such as blood alcohol levels, satellite images and on-site photography. Its sophisticated design enables it to evaluate both individual and combined risk factors for car crashes, delivering a nuanced understanding of how these elements interact.
The innovative tool’s continuous learning loop ensures that its predictive accuracy improves as more crash-related data is introduced. This unique feature makes the tool increasingly reliable over time.
The researchers can also quantify the trustworthiness of each prediction, providing a percentage accuracy for real-world scenarios.
“By reframing crash prediction as a reasoning task and using LLMs to integrate written and visual data, the stakeholders can move from coarse, aggregate statistics to a fine-tuned understanding of what causes specific crashes,” Yang added.
SafeTraffic Copilot aims to serve as an invaluable asset for policymakers and transportation designers, enabling them to identify risk factor combinations and implement evidence-based interventions. The tool’s insights can guide more effective infrastructure planning and safety measures, ultimately saving lives and reducing injuries on American roads.
“Rather than replacing humans, LLMs should serve as copilots — processing information, identifying patterns and quantifying risks — while humans remain the final decision-makers,” added Yang.
The team’s broader vision includes applying SafeTraffic Copilot as a template for responsible integration of AI in other high-stakes areas like public health and human safety.
Given the black-box nature of LLM-based models, the researchers continue to explore ways to make AI-driven decisions transparent, accountable and aligned with societal values.
“The central focus of our ongoing research is to find the best way to combine the strengths of humans and LLMs so that decisions in high-stakes domains are not only data-driven, but also transparent, accountable and aligned with societal values,” Yang added.
The study’s co-authors include Hongru Du, an assistant professor at the University of Virginia, along with Johns Hopkins doctoral candidates Yang Zhao, Pu Wang and Yibo Zhao.
Source: Johns Hopkins University