A video has been posted by Google’s Waymo self-driving car project to showcase how its driverless technology accounts for cyclists. In the footage, the vehicle predicts that two cyclists will move out to pass a vehicle blocking the cycle lane and slows to allow them to safely pull out.
Since 2009, Waymo vehicles have driven more than 10 million miles on the road, creating a vast library of the kinds of scenarios a driver might encounter.
The firm says its cars are programmed to recognise cyclists as ‘unique users of the road’ and drive conservatively around them. The cars can recognise common hand signals and have been taught to recognise other common behaviour as well.
In an article on Medium, Waymo’s Chief Safety Officer Deborah Hersman writes: “Safely sharing the road is an important part of driving, and the Waymo driver tirelessly scans for objects around the vehicle — including pedestrians, cyclists, vehicles, road workers, animals, and obstructions — and then predicts their future movements based on information such as speed, trajectory, and road context.”
Hersman says the vehicles are designed to understand that different types of road users will not behave in the same way.
“Our technology is able to perceive and accurately classify different road users which, in turn, allows our self driving vehicles to predict their behaviour.”
She continues: “As we work to build the world’s most experienced driver, we’re putting considerable thought and engineering into ensuring our vehicles can understand cyclists’ unique behaviour and are ready to act with their safety and protection top of mind.
“In this video, a Waymo vehicle sees a cyclist and predicts that the rider will shift out of the bike lane to get around a parked trailer. Our Waymo driver is designed to drive defensively around cyclists to help prevent the common collision scenarios that occur between cyclists and human drivers every day.”
One of the issues Waymo has faced is that quite often its vehicles drive too cautiously. The vehicles have been hit from behind on several occasions as a result of sudden, unexpected braking.
In 2015, a cyclist told of how he had confused one of the firm’s cars by trackstanding at a junction. The firm said it welcomed that incident, as these sorts of events help the software learn about bike riders' behaviour.
Shortly afterwards, Dmitri Dolgov, principal engineer of the project, employed somewhat ill-advised wording, saying that the firm was working to make its cars ‘more aggressive’.