Oxbotica unlocks the potential of driverless cars

As a young PhD student in 1996, Paul Newman was working on robotics and machine learning when a friend asked him a question that changed his life: “How can a robot know where it is without looking at a map?” He recalls thinking: “Wow! That’s the whole show. Everything is a derivative of that question.”

Two decades on, the technology that resulted from his search for an answer has been used in driverless taxis in Greenwich, south-east London, in submarine navigation and in Nasa’s Mars rover.

Newman runs Oxbotica, a spin-out from Oxford university that he co-founded with Ingmar Posner. They met as research students at Oxford and in 2003 decided to pool their efforts to develop a system that can recognise its location and surroundings.

Oxbotica’s main product is called Selenium, and it is not a driverless car as such but the robotic brain that can run one. The software can be applied to, in Newman’s words, “all things that move”, whether on distant planets or in local warehouses in autonomous forklift trucks.

“We never called ourselves a driverless car group,” says Posner. “We’re about fundamental research in mobile autonomy. That has applications in all places.” The company is in talks to supply makers of mining trucks and port operators, and plans to have fully self-driving cars on UK roads within 18 months in collaboration with a car manufacturer.

Because Oxbotica vehicles are connected to the same “brain”, or central learning system, they share experiences, benefiting from the observations and mistakes of others on the system. “The vehicles learn,” says Posner. “That is what differentiates our technology from almost all the other technology out there.”

The company is only three years old, having previously been part of the university’s Oxford Robotics Institute (ORI) research group, run by Newman and Posner. Almost all the engineers and technicians who helped it at ORI now work for Oxbotica and it continues to license some of the technology from ORI. It has taken no outside funding, surviving initially on research grants and, as an independent business, from its revenues.

The business is still too small to publish full financial statements but abridged accounts filed at Companies House show it made a profit of £478 in the 12 months to September 2015, its first year of trading.

When Newman and Posner formed a board, they brought in Graeme Smith as chief executive to run day-to-day business. He had led carmaker Ford’s telematics division, though is an inventor himself, with registered patents in adaptive cruise control.

Newman and Posner own just under 60 per cent of Oxbotica’s shares, while Oxford university has a 25 per cent stake. The rest is shared among the board, while employees are given share options when they join.

Expansion is in the pipeline: Newman and Posner are considering an initial public offering, sale or other form of fundraising. The company has 33 employees, aims to double this in the coming months and has just taken an office in Oxford that can accommodate 95 staff.

Our vehicles ‘learn’, which is what differentiates our technology

Its current home is at the UK Atomic Energy Authority, where it has a warehouse the size of a small aircraft hangar, as well as office space and meeting rooms. There is not a beanbag in sight, but then this is not the average start-up. Newman and Posner, sitting at a meeting table strewn with laptops, notebooks and coffee cups, talk of the unorthodox genesis of the company. Most new businesses are born out of a single idea the founders wish to turn into gold. “We came with 43 things,” says Posner. “It turns out a driverless car needs more than one ‘thing’.”

In its early days, the group shared offices with academics who did not care for robots whirring around them. Woolly hats from charity shops were put on the robots’ wheels to prevent damage to the carpet.

For several years the small research team needled away at software involved in such unglamorous areas as calibrating the different sensors in a vehicle. “All of that is very non-glitzy,” says Newman, adding that if such work is not done, it is “going to catch you out”.

After some years working with companies such as BAE and Nissan to test the feasibility of their self-driving capability, the pair were left thinking they would always be “running prototypes”. “We were still hungry,” Newman says.

The building where most of their research happens is only sparsely insulated against the winter chill outside. Several vehicles sit in various stages of disassembly, though any from established carmakers have been cleared away before my visit.

A golf cart-style buggy stands with its wired innards hanging out. It is being engineered for use in parcel deliveries. A pod used in the Greenwich project, a government-funded scheme to test public reaction to autonomous technology, is there, as well as the company’s flagship self-driving car, an adapted Renault Twizy compact electric vehicle called Genie.

At meetings with potential clients, one of the first things the company does is to show them a colour chart in the shape of a cube, with each corner marking a different ability required to make an autonomous vehicle, such as “mapping”, “perception” or “navigation”.

“If you mix all of those colours together you get the bright white of full-blown autonomy. That’s Selenium,” says Newman.

Of course, not all clients are looking for an all-singing, all-dancing self-driving system. Some just want sensor technology, or mapping, or location. Nasa’s requirement was for Oxbotica vision sensors for its Mars rover.

“Nasa is very conservative when it comes to putting technology on to those things,” notes Posner. Apparently, Nasa technicians balk at the idea of a self-driving robot that might break down 55m km from the nearest engineer.

Transport is bust and will get more bust unless we fix it

Newman began to develop autonomous technology to work in similarly hostile conditions, namely while programming submarines for a PhD project at the University of Sydney in Australia. Underwater is a “wildly hard place to start doing robotics, because it’s totally unforgiving”, he says.

The ability to pinpoint location remains the core question that drives everything Oxbotica pursues today. Newman’s code for the submarine remains in use, he says, explaining that a friend had emailed him recently to tell him the craft was under the North Pole.

On dry land, the implications of successfully deploying self-driving technology are vast. Carmakers and technology groups are developing autonomous vehicles as one way to reduce road deaths that result from human error.

Oxbotica has talked to almost all of them, has partnerships with some and will be launching ventures with others in the months to come, Newman says. But fully autonomous pods — such as those recently trialled by Oxbotica in Milton Keynes, central England, — also have the potential to ease urban congestion, improve mobility for elderly people and even kill off traditional car ownership.

“Transport is bust,” says Newman. “And it’s going to get more busted unless we do something to fix it. What we stand on the cusp of here is being able to contribute in the UK to changing what transport looks like in some places.”

There are several hurdles to overcome, however: public perceptions of safety and regulation are almost as formidable as barriers as any technological challenge.

During the Milton Keynes project, for example, Oxbotica was surprised by pedestrians’ reactions. Once they had learnt that the vehicles were programmed to avoid hitting them, several adopted somewhat cavalier attitudes towards the cars.

“They just didn’t look,” says Newman. “One person ran and jumped in front of the car, failing to take account of the stopping distance required. People assume that because there’s no driver, the vehicle has no mass, but one tonne of metal still moves independently of the machine inside it saying ‘Oh dear, someone’s in front of me’.”

Another concern is that if other drivers know these cars are designed with safety in mind, many will take advantage of that to cut them up or refuse to let them out at junctions. The solution, Newman and Posner believe, is the machine learning part of Selenium, where cars become more deeply embedded in their local environment after learning from every journey. A vehicle in New York, for instance, would learn to push forwards like local taxi drivers.

But Newman and Posner have some serious rivals. Google, Baidu and other technology titans are working on similar systems, so how can a British minnow compete? Newman says negative views on the subject would be like telling Dell in the 1980s that it should not make computers because IBM already existed.

“You’ve got to get in the race,” he says. “And that’s where we are.”