On any given day, the average person’s biggest chance of death or serious injury comes in one place: On the road. Could AI play a role in fixing that?
“I have no doubts that AI will help make us safer on the road,” says Prof. Yuanchang Xie of the Department of Civil and Environmental Engineering.
For starters, AI can be used to analyze historical crash data and identify hidden and subtle patterns, he says, explaining that this helps transportation agencies improve roadway design, safety and maintenance.
“I use AI extensively in my research,” says Xie, an expert on transportation engineering. Last year, Xie and his co-researchers from UMass Amherst were awarded a $100,000 grant by the Massachusetts Department of Transportation to use AI to detect pedestrian crossings from aerial images for the entire state.
Xie was also a co-principal investigator in two previous research projects funded by the National Science Foundation: the ethics of self-driving cars and the impact of automated vehicles on traffic flow. He says the safety of self-driving cars will continue to be improved with the help of AI.
“For instance, AI, coupled with advanced lidar, radar and/or camera sensors, can detect surrounding vulnerable road users, such as pedestrians, joggers and cyclists, and alert drivers” to their presence, he says. “This is particularly useful in low-light conditions, in which human drivers may struggle to see what is ahead of them.”
AI can also analyze driver-behavior data captured using onboard diagnostics systems to identify fatigued driving and erratic behavior for driver training and education, he says.
“AI also plays an increasingly vital role in processing data captured by roadside sensors to extract trajectories of all road users. These trajectories help identify dangerous conflicts or near-crash events not reflected in historical crash reports,” he says. “This information is crucial for transportation agencies to develop pro-active safety improvement strategies at places like intersections and exit ramps.”
Although there is much to be optimistic about when it comes to AI and road safety, Xie says AI models have limitations resulting from their training samples.
“They can make mistakes,” he says, when faced with scenarios not represented in their training datasets.
“These cases are a primary reason for the limited deployment of fully auto-mated cars on the road,” Xie says.—EA