Driverless cars offer plenty of advantages: autonomy for those who cannot drive themselves, decreased pollution and better traffic management.
But while companies such as Uber are committed to putting driverless cars on the road, they’re too costly an option for mass transit. Driverless cars rely on light detection and ranging (LiDAR) sensors to handle the complex tasks of understanding what is taking place around them, planning a path and avoiding obstacles. And those LIDAR sensors add tens of thousands of dollars to the cost of a car.
AI researcher Raquel Urtasun, head of Uber Advanced Technologies Group (ATG) Toronto, associate professor at the University of Toronto and founding member of research organization the Vector Institute, is pioneering technology that she thinks can replace LiDAR. She’s combining computer vision software, sensors and machine learning to help the driverless car “see” its environment. This method takes in data from the car’s surroundings to help predict the actions of other elements of traffic.
Urtasun has been able to show results similar to those achieved with LiDAR sensors using this less-expensive technology. And LiDAR doesn’t need to be entirely replaced to have an economic impact. Simply reducing the number of LiDAR sensors a vehicle needs reduces its cost.
Urtasun’s technology can also help develop the high-resolution maps a driverless car requires to navigate. It helps the car localize — or pinpoint its location — replacing the need for complex mapping, another potential way to reduce the cost of putting large numbers of driverless vehicles on the road.
While the increased availability of driverless cars can boost mobility and ease traffic congestion, Urtasun cites another critical reason for pursuing this technology: improved safety. “More than 1 million people a year die due to traffic accidents,” she says. “Imagine the potential impact that we can have if we can reduce that number significantly.”