That day in the future when you get inside a self-driving bus to get to work, you’ll assume that the bus is programmed to drive you there as efficiently and safely as possible. Software will have replaced the driver who does that today. Perhaps more importantly, software will have also replaced those managing scores of busses at a control centre. Robin North is building the latter technology that will control those fleets from the mothership.
North and three other engineers spun out Immense Simulations from their British-government funded research group and are selling software that can coordinate thousands of driverless vehicles at once. Such cars will be “strategically autonomous” he says, because they’ll be synched up as part of a huge network.
Immense is targeting traditional fleet operators, city authorities like Transport for London and other new technology companies keen to enter the fledgling market.
Its software (pictured below) runs on a cloud-based, simulation platform from London startup Improbable. The platform can host large-scale, detailed simulations of cities, which allows Immense to simulate how cars would react to changes in the weather, pedestrians or other cars on the road.
Immense was one of the first companies to work with Improbable last year when the latter company first introduced its Spatial OS platform to early clients, many of whom were game developers.
“We were some of the first people through the door,” says North. “We set ourselves a challenge of, ‘Ok, you can do a big, scaleable, virtual world. Can you do a real one with millions of entities? What happens if you try and simulate Manchester?’”
North’s team pulled together a proof of concept in the summer of 2015, back then under the tutelage of the British government’s Transport System’s Catapult network. The network has 10 ‘innovation centers’ across the country aimed at bringing academic research into technology companies.
Much of the math at the core of Immense Simulations revolves around how best to organize a fleet of hundreds or even thousands of vehicles, making sure that each is going to the right destination to make the network as efficient as possible.
It involves solving the last-mile issue, a perennial problem among delivery operators in getting from port to the final destination while bumping into things like road works, cyclists and skateboarders. North calls this the “messy business of operational management.”
Uber has its own routing engine that, in a more basic way, finds the most efficient route between human drivers and passengers.
Called Gurafu, Uber built the engine in March 2014 to help it fix the problem of “cold starts” in new cities, when its algorithm struggled to find the most efficient routes for drivers because it hadn’t yet built up a familiarity with the area. It’s a highly-complex mathematical process built on well-known search algorithms and contraction hierarchies that makes a driver’s ETA more accurate and quicker to calculate.
The work of modelling a fleet of drivers is never done, though. Uber and other modern-day fleet operators are constantly asking the question of where to send a driver when they’re not on a job, or how to better shave off a driver’s waiting times. And it’ll take on a whole new meaning when autonomous cars come into play.
“This stuff is going to be really important to companies like Uber and Lyft in the future,” says Herman Narula, who co-founded Improbable. “It’s not enough to get the cars driving themselves. You need to think about the problem of large-scale fleets.”
In early 2015 Uber poached 34 engineers and six managers from Carnegie Mellon’s National Robotic Centre in Pittsburg in what was widely seen as a step towards developing its own driverless cars.
Autonomous vehicles are still years away, but by 2035 they’ll represent about 10% of light-vehicle sales around the world, according to a recent study by
Narula says Improbable has received “a huge set of enquiries” from companies about managing driverless fleets and predicting their impact on the world around them.
Insurers, for instance, have no idea how to price premiums in a world that has both autonomous and non-autonomous cars driving the roads. Some have asked Improbable to run simulations to work out some initial pricing models.
“It’s 10 or 15 years away,” North says of autonomous fleets of cabs, buses and delivery vehicles. “But there’s going to be niche applications and a lot will be on the road sooner than that.”
North points to some of the trials taking place globally, including 15-mph driverless pods in Milton Keynes, England and Singapore’s testing of self-driving shuttle vehicles. “We’re looking at things that are five or 10 years away from deployment.”
The driverless pod in Milton Keynes, known as the LUTZ Pathfinder, has its own autonomous control software developed by the Mobile Robotics Group in Oxford University.
But studies show that autonomous fleets of cars — as opposed to driverless cars that operate independently — could cut down on congestion in cities enormously.
Recent research in Lisbon, Portugal showed that shared, driverless fleets of cars could meet the city’s mobility needs with just 35% of the usual number of vehicles on the roads during peak hours.
For a full 24 hours on average, they city would only need 10% of existing cars on the road to meet its transportation needs, according to the 2015 study on “TaxiBots” by the International Transport Forum, a division of rich-country think tank the OECD.
Ask North about his future customers and he can’t give concrete answers: they could be traditional fleet operators or new, technology-based fleet operators of the future. He says city authorities like Transport for London are “really interested.”
But as for those crucial, first customers: “We don’t know who they’re doing to be yet,” he says. “But we know the things they’re going to need to do.”