Informatics Professor Applies AI to Auto Traffic Simulation
Assistant Professor Hua Wei is studying how to build more realistic models for traffic simulation, with the goal of improving predictions by closing the gap between just following traffic laws vs. how people actually drive.
That gap can be significant, so Wei is using the latest in artificial intelligence and reinforcement learning to help his cause, supported by a $175,000 research grant from the National Science Foundation.
Wei explained that reinforcement learning has seen success in artificial intelligence domains such as virtual gaming where simulations are constructed using unlimited yet hypothetical data to achieve specific results. However, despite being effective in certain predetermined learning goals, it has rarely proven so when applied to real-world physical systems such as traffic. That's because traffic is an environment where the influence of human behavior is significant, but data is sparse and difficult to obtain, Wei explained.
The investigation following the infamous "Miracle on the Hudson" incident, where a U.S. Airways flight was forced to emergency land in the Hudson River after a double bird strike to the engines, is a well-known example of how a simulator can overlook the human element.
The simulators the National Transportation Safety Board used to investigate the incident cast doubt on the pilots’ decision for a water landing instead of trying to navigate to a nearby airport. Only when investigators realized that their simulation was unrealistic did they conclude that the pilots had made the correct decision. "The immediate turn made by the pilots during the simulations did not reflect or account for real-world considerations, such as the time delay required to recognize the bird strike and decide on a course of action,” their report stated.
Wei will begin by testing his new computer simulations with a single driver and progress to models incorporating multiple drivers. Though focusing on human drivers, Wei’s project could benefit all modes of transportation. Wei will present his findings to the U.S. Department of Transportation as well as Lawrence Berkeley National Laboratory.
The research is also notable because Wei joined NJIT's Ying Wu College of Computing, and the Department of Informatics, just two months ago. “This is exciting. I am in the early stages of my career, and to be recognized by a federal agency such as the NSF during my first year as a professor is beyond my expectations,” he said.
When asked why he chose to focus on the issue of traffic efficiency and safety, the answer was based on personal experience — “I missed my high school entrance exam in physics because my parents were stuck in an incredible traffic jam in China!” Wei has been determined to use his skill and expertise to address this global problem ever since.