Researchers at the University of Waterloo will spend the next year attempting to improve its autonomous vehicle’s ability to identify pedestrians and predict their path of travel.
Human identification is one of the more difficult tasks for a driverless vehicle, said Ross McKenzie, the director of the Waterloo Centre for Automotive Research.
“We’ve done some work with pedestrians, and there are significant variations there and challenges,” he said.
For example, a group of people looks much different than a solitary person. Sometimes, people are pushing shopping carts, using a walker or walking their bike across the street.
“We can identify a pedestrian with a reasonable degree of accuracy, but then these variations of people clustered together or people holding something or walking with something, are the variables we need to teach the software to recognize.
“The technology is there to detect the object, but the challenge is object recognition.”
‘Difficult to avoid’
Sunday evening, an autonomous Uber vehicle, a Volvo XC90 operating in autonomous mode with an operator behind the wheel, collided with Elaine Herzberg, 49, who was walking her bicycle across a four-lane road in the Phoenix suburb of Tempe. The Uber vehicle traveling at about 40 mph, police said. Herzberg died of her injuries.
Tempe Police Chief Sylvia Moir told the San Francisco Chronicle that from viewing videos taken from the vehicle “it’s very clear it would have been difficult to avoid this collision in any kind of mode [autonomous or human-driven] based on how she came from the shadows right into the roadway.”
Uber immediately stopped all AV pilots on public roads, including one in Toronto. Toyota Motor Corp. followed suit.
“What it will do across the segment of autonomous-vehicle development, it will cause all of us to be even more vigilant,” McKenzie said of the Arizona incident. “We’ve erred on the side of safety from the outset.”
The University of Waterloo has conducted just one test on public roads in Ontario, McKenzie said. That took place in December 2017, when researchers needed snow-covered roads for a specific test. Otherwise, research is conducted on closed courses.
Blackberry QNX, a Canadian company which also conducts autonomous vehicle testing in Ontario, but does so on closed courses for safety reasons.
Safety first in Ontario
Ontario’s Automated Vehicle Program that allows companies and researchers to test autonomous technology in real-world situations on public roads. A government-issued permit is required.
Ministry of Ontario spokesman Bob Nichols said all companies testing autonomous technology on Ontario roads “are expected to follow a strict set of guidelines and rules, designed to provide the utmost safety for all road users.”
In 2017, suppliers Magna and Continental started an autonomous vehicle test in Detroit. Each drove autonomous vehicles from Detroit to Windsor and then to Sarnia, before crossing back into Michigan.
Magna said that “at this time it would premature to comment” about how the incident in Arizona might affect research moving forward.
Spokeswoman Tracy Fuerst said the collision was “certainly unfortunate and our thoughts are with all of those involved.”
Nichols said the province will be following the situation in Arizona closely “and will consider what measures are appropriate as more becomes known.”
Meanwhile, research at the University of Waterloo will continue, too.
“Ultimately, all the work we’re doing is driven by the idea that computer-operated autonomous vehicles will be safer than a human-operated vehicle,” McKenzie said. “With a human-operated vehicle, you can have somebody fall asleep at the wheel; you can have someone leave late and be a little more aggressive behind the wheel. Autonomous vehicles are conservative by nature and do not take chances.
“As far as our approach here, we continue with the same objective, which is to develop a vehicle that is more reliable and safer to operate than a human-operated vehicle.”