But cities have some urgent questions to answer, and failure to address the issues raised could see us sleepwalking back into the problems of the 1960s and 70s, where cities became thoroughfares for traffic first … and places for people second.
The ‘problem’ posed by cyclists
Driverless cars navigate and detect other road users using a combination of cameras, detailed maps, radar and, in the case of Google cars, Lidar (light detection and ranging), a laser-sensing system adapted from oceanographic surveying. Google, in a company now spun off as Waymo, has been testing driverless cars (with pilots inside) on public streets in the US since 2009, clocking 2.5 million miles, and honing the technology following interactions with other road users.
A driverless car will, in theory, stop if it detects an object in its path – but cyclists, being small and agile, represent a unique challenge. AVs struggle with changes in speed and the huge variety of cycle shapes and sizes. They even struggle to detect which way a bicycle is pointing. Deep3DBox, a program designed to identify 3D objects from 2D images, such as camera footage, is the most successful at doing this; yet it only spots a cyclist in 74% of cases, and correctly predicts the direction they are facing just 59% of the time. Poor weather makes detection even less accurate.
Former Renault-Nissan chief executive Carlos Ghosn described cyclists as “one of the biggest problems for driverless cars” last year. They confuse the vehicles, he said, because at times they behave like pedestrians, at other times like cyclists, and “they don’t respect any rules usually”.
Google has acknowledged that “it’s hard for others to anticipate their movements”. This came after one cyclist bamboozled a self-driving Lexus by performing a prolonged track stand at a junction. Google has since taught its cars to recognize cyclists’ hand signals, different sizes and shapes of bike, and allows them more space on the road.
The issue of detecting and reacting to unpredictable behaviour is far from solved, though, as the Guardian recently witnessed during a ride in a driverless Nissan Leaf. In a separate incident earlier this year a driverless Leaf was caught on camera overtaking a cyclist at very close proximity, even though the vehicle’s monitors indicated it had detected the rider.
‘Unworkable’
But what action should a driverless car be programed to take when it sees a cyclist or a pedestrian in its path? And what happens if people crossing roads learn they can simply walk in front of AVs which will be forced to brake?
Robin Hickman, a reader in transport and city planning at University College London’s Bartlett School of Planning, believes this makes driverless cars “unworkable” on busy urban streets.
“In terms of the algorithm for dealing with obstacles that move in unpredictable ways, like cyclists or pedestrians, I would say that’s unsolvable,” says Hickman. “If a pedestrian knows it’s an automated vehicle, they will just take the priority. It would take you hours to drive down a street in any urban area.
Read more at Street Wars 2035: Can Cyclists and Driverless Cars Ever Co-Exist?
No comments:
Post a Comment