New laser system could let driverless cars 'look around corners'

It is a parent’s worst nightmare – their child’s ball rolls into the road and into the path of a car, putting the child at risk as it runs after it.

But even though the ball is around a corner and out of sight of people in the car, the vehicle quickly brakes and rolls to a stop. That is the scenario picked by Stanford University researchers to highlight their new work, which they said could one day make cars safer.

“It sounds like magic, but the idea of non-line-of-sight imaging is actually feasible,” said electrical engineer Gordon Wetzstein. The Stanford team is just one group of researchers bouncing lasers around corners to capture images of objects, a system which could benefit autonomous and semi-autonomous cars.

The researchers put a laser next to a highly-sensitive photon detector, which can record single particles of light. They shot laser pulses at a wall, which then reflected onto objects around a corner. The lasers then bounced off the object and back to the wall, then the detector.

An algorithm which quickly reconstructed the reflected photons into an image is what sets the work apart, the team said. In a video shared with journalists, the system created a partial 3D scan of a small wooden mannequin, showing a distinct head, arms, legs and torso.

A similar system could help autonomous cars make quicker and better decisions, the researchers said. “This is a big step forward for our field that will hopefully benefit all of us,” said senior author Wetzstein. “In the future, we want to make it even more practical 'in the wild.'”

The technology could potentially make use of existing Lidar (Light Detection and Ranging) systems currently installed on cars, the Stanford engineers said, although the imaging technology might need adapting as it currently only focuses on directly-reflected rays.

Outside the visual field

A system helping cars look around corners could be a useful equivalent to human drivers’ hearing, said RAC chief engineer David Bizley. “Clearly there will be situations where having knowledge of what is going on outside the visual field will be useful,” he said to Professional Engineering. “There are situations where drivers get some clues as to what is happening in areas they can’t see, and this clearly is helpful in providing an alternative.”

However, Bizley said it will take more than a technological development to convince the public that driverless cars are safe. Last year, an IMechE and ICM Unlimited report found that 66% of people would feel uncomfortable travelling in an autonomous vehicle with no human intervention.

 “It is a far more emotional response that is leaving people concerned,” said Bizley. “I don’t think anything other than a track record of safety will do it.” People also simply enjoy driving, he added.

Practical use

Despite the Stanford team’s success, researchers admitted the technology still needs significant development. Although their algorithm successfully reconstructed images in less than a second, scans still take between two minutes and an hour – far too long for practical use – depending on lighting and the hidden object’s reflectivity. The engineers also hope to upgrade the system to visualise moving objects, like a bouncing ball or running child.

The research was published in Nature.