Oxbotica

Intelligent Machines

One way to get self-driving cars on the road faster: let insurers control them

Data gathered by autonomous cars and shared with insurance companies could be used to keep the vehicles from taking undue risks.

At this moment there are hundreds of autonomous cars on the roads around the world. They’re all still experimental (we’re not counting Teslas, which are not autonomous), because we don’t really know how safe they are yet. And, importantly, if we were to turn them loose carrying people around and they ended up in a crash, we wouldn’t know who to blame.

This question of liability is one of the thornier issues surrounding autonomous vehicles, especially if they’re ever to be sold to the public. But it could be solved with help from a trial led by Oxbotica, the autonomous-vehicle company spun out from the University of Oxford, to investigate how driverless vehicles can gather and share data. The idea is to explore not just how cars could pass data between each other in order to drive more effectively, but also how that data could be used by third parties like municipal authorities and, crucially, insurers.

The project uses three autonomous Ford Fusions, each loaded with stereo cameras and lidar sensors and controlled using Oxbotica’s autonomy software, Selenium, as they drive themselves around Oxford and the surrounding countryside. All the vehicles use cellular connections to send data to each other and to other organizations that are involved in the project, such as insurer XL Catlin.

The idea of wirelessly gathering data from cars isn’t new, but autonomous vehicles collect an incredible amount of it as they navigate the world—in the case of the Oxbotica cars, terabytes’ worth in just a single day of driving. That means figuring out what, exactly, to share is a difficult problem, particularly given that the data must all be transmitted via cellular connections.

There are, of course, security concerns about connecting an autonomous car to the internet. Waymo’s CEO, John Krafcik, has said that its cars go online “only when they need to, so there isn’t a continuous line that’s able to be hacked, going into the car.”

For now, Oxbotica’s vehicles only transmit high-level information such as speed, direction, and inferences about their surroundings. Paul Newman, a professor at Oxford and cofounder of Oxbotica, says the cars are already getting help planning their routes by alerting one another to stretches of road that have changed since previous trips—for example, when construction work has started overnight.

More interesting, and fraught, is the prospect of sharing data with third parties. “When these vehicles are running, they’re doing mathematics—the outputs of which are probabilities about things, on which decisions are made,” explains Newman. “How sure are they of their position on the road, the color of a traffic signal, or the speed of an oncoming vehicle?”

Oxbotica already shares some of that information with XL Catlin—how many geographic features the car recognizes, say, or how many obstacles there are nearby—in order to create risk scores that could be used to determine how the car should behave.

Newman gives a hypothetical: say a car spots a large group of children on the sidewalk, near a school, in the middle of the afternoon. While it doesn’t understand that class is letting out for the day, it does see more potential obstacles than usual. An insurer could process that data and then allow robotic vehicles down the road only at a low speed, or else have them re-routed.

“Insurers can adjust the envelope [in which a car can operate] to control the risk on the policy,” explains Newman. “The autonomy system has insurance built into it that allows it to control risk over a fleet.”

Oxbotica reckons that this kind of close relationship with insurers could help encourage lawmakers to allow more autonomous cars on the roads for testing, buoyed by the knowledge that expert risk assessors are involved in controlling them. Newman also suggests that this could make it easier to get the go-ahead to test autonomous vehicles in other high-stakes environments, like airports.

“If insurers are willing to account for the uncertainties, then it may speed up testing,” says Jack Stilgoe, a senior lecturer at University College London who specializes in the governance of emerging technologies.

But he adds that insurance is a relatively narrow issue in the face of the technology’s potential to change how we get around. And insurers, if left to decide when and how autonomous cars are deployed, might not realize that potential.

“The insurance industry won’t be thinking about such things. Policymakers need to,” he says. “There is an opportunity to rethink transport and cities.”