From the start, the team at DeepMind wanted to build a vehicle that would work on its own, on a road that could be paved with gold.
That meant making sure it could take the right decisions.
“When you’re driving around a town and you see a sign saying ‘no cars’ or ‘no pedestrians’, that’s a big red flag that the system can’t be trusted,” says Martin Rees, a DeepMind engineer who joined the team in 2011.
“We need to make sure that’s the case in all cases, whether it’s pedestrians or cars.”
That meant a system that could understand its surroundings, recognise other vehicles, and recognise that the signs in the middle of the street meant pedestrians could pass without fear.
“Our first thing was to get the signal system to be a good enough signal for the system to understand that there’s a pedestrian in front of us,” says Rees.
This is the key to making the system as safe as possible.
This image shows a self-balancing vehicle at a pedestrian crossing.
This picture shows the system driving over the road.
That’s where the AI-powered car comes in.
A new class of driverless car The first step was to build something that could follow a pedestrian across a road.
In the first test run, the system got stuck, and had to be replaced.
It took three days to fix it.
“I didn’t want to make this system as hard to get into as possible,” says DeepMind’s Rees in the film.
“But the best way to make it as hard as possible was to make the system be able to read the road, which is what you need to be able [to] navigate a street in a city.”
The final result, as shown in the trailer, was a self‑balancing car that could navigate the road without human assistance.
The AI system could follow the road and avoid obstacles.
But it needed to be programmed to make that happen, in order to avoid accidents.
“The reason it took so long to build the signal systems is because it’s really difficult to get a signal system that is good enough for a city road,” says Alex van der Kolk, the director of the Oxford Centre for Transport at the University of Oxford.
“To build the system for a road, you need a good system, and you need good signals.
You need the right kind of signals for the right conditions.
The signal systems we have today are so far from those conditions.”
The next step was building a system to use the signals in a more intelligent way.
“It’s not a simple task to build an AI system that understands the road signals,” says van der kolk.
“If you build it in a way that’s easy to build it, then you’ll be able build it well.”
The team built a system called “Sonic” that can follow a car, or a car in front.
This system can avoid obstacles and pass in front if there is a pedestrian on the right.
This new system can follow pedestrians without human intervention.
This car has been programmed to drive over a pedestrian crosswalk, to make a left turn and cross the road safely.
But if a pedestrian tries to run over the car, the car will turn into a wall of red lights.
This means the car can’t stop, and the driver must avoid getting hit by pedestrians.
The system is programmed to follow a “pedestrian crosswalk” on the road as it follows a pedestrian.
This green box shows the signal that the car uses.
The red box shows a warning light that the signal is not good enough.
“Once you start to put that information into the system, it becomes much easier to understand what is the right signal,” van der,kolk says.
This video shows how the system understands pedestrians on the street.
This next video shows the car turning into a pedestrian “walkway”.
The system can stop at the signal and turn into an “alternative” path for a pedestrian, which has no red light.
This design allows the system “to know the signals of other vehicles that are following and then pass through the signal safely,” van denkol says.
“You can imagine this as a pedestrian walkway in a car park.”
The first car that drove over the pedestrian crosswalks was a prototype in March 2016, but the team has not built it.
In 2018, the same team had its first success when it successfully drove over a walkway on the London Overground.
This vehicle could follow pedestrians, even if it was not on the crosswalk.
But the new system has a lot of improvements.
It can now navigate the city with less human intervention, and is programmed so it can do so without any human intervention at all.
This diagram shows the traffic signal system, with green boxes showing pedestrians on one side, and red boxes showing cars on the other.
The new system is being built by an artificial intelligence company called