Helping to steer self-driving vehicles in the right direction

How much of what a driverless car “sees” is actually vital to the process of driving — and how do you “train” the car to cherry-pick the correct information? Trained drivers quickly learn to spot danger signs; a flash of brake light or a pedestrian dashing towards a kerb, but also to weed out the irrelevant, such as distant trees or buildings or the puff of smoke from an exhaust pipe.

Subliminally, experienced drivers respond to or discard these cues, reacting automatically, slowing or manoeuvring accordingly.

A car’s under-bonnet brain, however, can become overloaded with the never-ending flood of data from its sensors; how does it know if that puff of smoke is just gas — or a solid obstacle?

Processing excess irrelevant data could trigger meltdown. But processing too little could mean vital clues — such as a football bouncing across the road — being missed. Which is why a consortium, MOVE_UK, has been conducting trials, right under our noses, backed by a £3.4 million grant from the Government’s Intelligent Mobility Fund. Royal Borough Of Greenwich council workers have been driving specially equipped Land Rover Discoveries, which have recently clocked up 30,000 test miles, to help streamline this complex task.

Each is fitted with cameras and sophisticated under-bonnet computers capturing a flow of information from the road ahead — equal to downloading a Netflix movie every few seconds. It is then beamed to a central cloud, allowing researchers to analyse it remotely, using newly developed tools.

They can then establish how an autonomous car would have reacted to hazards and compare this with how the human driver reacted in real world conditions, enabling the autonomous ‘brain’ to learn and evolve. Researchers can spot which hazards the car ‘observed’ — which it missed.

This enables the consortium — Bosch, TRL, Jaguar Land Rover, Direct Line, analysts at The Floow and Digital Greenwich — to refine how automated driving functions will eventually respond in real life, a key step towards self-driving cars appearing in the showroom. It will, they say, ensure that autonomous vehicles drive in a natural way, closer to that seasoned pro than a learner.

Phase one is over. The next two phases will see additional sensors fitted to the cars so that after three years, data will be gathered — and refined — from 360-degree sensors.

“The data collected is particularly valuable as it is being generated through ‘real world’ driving, rather than from the test track,” says Arun Srinivasan, executive vice-president and head of mobility solutions, Bosch UK. “The new validation method takes us one step closer to fully autonomous driving — and to our vision of accident-free and stress-free driving for the future.”

There are benefits for road-users and Greenwich, too, which is discovering how to streamline street infrastructure so that they help rather than hinder self-drive cars. It could, says Trevor Dorling, Digital Greenwich MD, lead to less cluttered roads, benefiting us all.

“It is so important that cities are at the heart of this debate,” he says.