Computational tools for society’s most complex challenges

Cathy Wu, a professor in MIT's Department of Civil and Environmental Engineering and the Institute for Data, Systems, and Society, has dedicated her career to solving complex problems to improve people's lives. Growing up in a family with a tight budget and her father's daily commute, Wu spent much of her childhood playing computer games, particularly "SimCity," which sparked her interest in designing efficient transportation systems.

Wu credits her older sister for instilling in her a desire to make a positive impact on people's lives.

Wu's journey in transportation research began during her undergraduate studies at MIT, after attending a lecture on autonomous vehicles by late professor Seth Teller. This lecture inspired Wu to work with Teller and later collaborate with Daniela Rus. Wu's research focuses on using machine learning and reinforcement learning (RL) to design safe, efficient transportation systems.

RL has the potential to exponentially improve the efficiency of transportation system design by modeling and analyzing multiple variants simultaneously, a task that traditional evidence-based approaches struggle with.

In 2018, Wu applied RL to analyze the potential traffic flow impact of autonomous vehicles in various traffic networks. Although RL initially yielded promising results, Wu faced challenges when trying to apply the technique to traffic problems in subsequent years. She eventually discovered that RL algorithms are highly sensitive, leading to a significant breakthrough in 2023.

Wu and her team devised a method to improve training efficiency by up to 30 times by focusing on the 10 percent of problems that RL solves well, collectively performing well on related problems.

Recently, Wu's research has focused on eco-driving measures, which involve intelligently controlling vehicle speeds to reduce excessive stopping and starting. Her work demonstrates that such measures could reduce vehicle emissions by 11 to 22 percent. This evidence supports the implementation of policies that could significantly improve transportation system efficiency, showcasing the potential of RL in informing transportation policy and solving practical problems.