Solar glare is making it tough for some self-driving cars being tested in Boston to read whether traffic lights are red, yellow, or green.
MIT spinout NuTonomy said that on some recent occasions, the “low evening sun and solar glare” have occasionally kept its autonomous vehicles from being able to read the color on traffic signals.
“In a sense, the challenge for an AV’s sensors resembles the challenge for human drivers: it can be difficult to perceive the state of a traffic light while staring into the sun,” the startup said in a quarterly operations update to Boston transportation officials.
The company explained its self-driving systems are equipped to keep the cars from blowing through a red light because they assume every traffic signal is red until it confirms there’s a green light. And if there’s a safety risk in waiting for that to happen, human drivers take over.
The issue adds another challenge to the list of human-simple, machine-difficult driving skills that companies developing autonomous vehicles still have to overcome. (People can put on sunglasses or use a sun visor to manage the blinding sun. Machines need an engineered solution.) The complexity of seeing traffic signals and interpreting how to respond has long been an issue for computers. For example, some have read brake signals as red traffic lights and others have seen patterns of leaves on trees as green lights, according to researchers.
To counter the glare issue, NuTonomy said it has started enhancing its computer vision algorithms by collecting low-light data that can pick up colors from the traffic signals.
“We have also made hardware adjustments, such as changing the exposure of a camera or adding glare shields,” the startup said in the update. “In the meantime, we have trained our safety drivers to be aware of this issue and know when to take over manual control preemptively.”
NuTonomy, owned by automotive technology company Aptiv (NYSE: APTV), has had permission from Boston authorities to operate its autonomous vehicles throughout the entire city since June 2018. That said, its cars stuck to their familiar routes during the fourth quarter of last year, running in autonomous mode on certain streets in the Seaport neighborhood and “in the periphery of South Boston,” the company said. It also ran its cars in manual (human-driven mode) throughout the Seaport and South Boston to collect data for mapping purposes.
No cars were involved in any collisions during the testing in the quarter, and there were no unanticipated “failures or disruptions” in the same time period, the company said.
During the fourth quarter, human drivers in the vehicles occasionally took over control of the AVs when emergency vehicles were nearby, when police were directing traffic, when construction vehicles obstructed the lane of travel, when oncoming vehicles or bicycles violated lane boundaries, or when other vehicles were “exhibiting erratic behavior.”
NuTonomy claimed that, with improvements in its software and additional technology development, its cars will be able to handle those situations without a human driver. It didn’t share a projected timeline.