Waterloo’s driverless vehicle hits 100-kilometre milestone
Software developed by UW is ranked among the best for detecting cars, people and objects
WATERLOO — A driverless vehicle under development at the University of Waterloo, dubbed the Autonomoose by grad students, has logged more than 100 kilometres on public roads in full autonomous mode.
“We hit the 100-kilometre milestone earlier this week, on Monday to be precise,” said Ross McKenzie, director of the university’s centre for automotive research.
Since December, a Lincoln MKZ equipped with nine cameras, lasers, radar and computers moved along the roads around King Street North and Northfield Drive in Waterloo. The car is hard to miss. It looks a like a giant, black bug with thick antennae sprouting from the roof.
Full autonomous mode means the car was driving itself and did not require a person to take over the wheel.
“It validates what we have done, and proves that we are in the game,” said McKenzie. “We wanted to make a vehicle operate autonomously on a Canadian road in all four seasons. We have done that.”
In fact, the software developed by the UW team is consistently ranked among the top three in the world by the Karlsruhe Institute of Technology in Germany for detecting cars, pedestrians, bicycles, motorcycles and stationary objects like signs, posts and curbs.
“The other two parties in the top three with us are Apple and Velodyne,” said McKenzie.
Velodyne is a leading manufacturer of laser and radar units for autonomous vehicles.
The Karlsruhe Institute posts a video recorded by a camera on the roof of a car as it travels on downtown streets. Driverless vehicle researchers around the world use the video to test their software. The institute does this to help set benchmarks that will lead to international standards for this emerging technology.
“Of all the participants, we are the only one making our software code publicly available,” said McKenzie. “It just fosters collaboration. It will encourage additional work by others. It could lead to partnerships with new entrants or existing players in the industry.”
McKenzie and his team use the Emergency Services Training and Research Centre on Erbs Road in Waterloo. The site has roads with pavement markings and an intersection with signals. So it is ideal for testing the Autonomoose. Government ministers used the location for their announcement in November 2016 that driverless vehicles could be tested on public roadways around the province.
For more than a year after that announcement, UW researchers developed software and algorithms for their autonomous vehicle, and tested the Autonomoose at the training site. It was taken on a public road for the first time last December.
The technology has to detect vehicles and pedestrians, know if they are moving, and accurately predict their pathway. It is doubly hard for pedestrians because people can move 90 degrees without changing location, or turn 180 degrees and move back. While making a 180 degree turn, a pedestrian appears to be stationary to the software navigating the car. Software developers need to work on that more so the car knows someone is about to move across the vehicle’s path again.
The laser and radar units on the vehicle constantly scan the road ahead, but could not tell the difference between patches of black ice and a pothole. So the software was enhanced to ask a series of questions when it detects a pothole. What month is it? What is the temperature? What has the weather been like during the previous 24 hours? Now, the Autonomoose knows the difference between a treacherous patch of slippery road and a hole in the asphalt.
“We are doing some unique things with ability to test in four full seasons,” said McKenzie.
As the Autonomoose is manually driven, cameras record everything around it. Back in the university lab that information is used to make a high-definition 3D map the software uses to navigate the car.
Right now, individual researchers and companies make their own 3D maps. Before autonomous vehicles are widely adopted those 3D maps must be readily available, said McKenzie.
He predicts driverless cars will be moving people around local streets within 20 years, but it will probably take 30 years before autonomous vehicles can safely navigate a trip to Toronto Pearson International Airport.
General Motors has Super Cruise on some 2018 Cadillacs for hands-free driving on Interstates and 400-series highways where there are no pedestrians, traffic signals or stop signs.
McKenzie said the Autonomoose software must recognize buses and big trucks, which is very difficult for developers at this point. City buses, school buses, double-decker GO buses and Greyhound buses have different silhouettes. A flatbed truck may or may not have a load on it. A flatbed tow truck could be hauling a sedan or a minivan.
“You can’t even predict what the silhouette will look like with a load, because loads can vary,” said McKenzie.