AI race car vs. human Who wins

By: Matthew Knight

Artificial intelligence has defeated world chess champions, mastered the ancient (and fiendishly complex) Chinese board game Go and more recently beat a bunch of professional poker players hands down.

But can a autonomous car go faster than a human driver on a race track?
Earlier this month, CNN’s Supercharged presenter Nicki Shields got the chance to find out as she pitted her driving skills against the world’s first autonomous electric racing car in Hong Kong.
Shields was racing in, as well as against, the “DevBot,” a prototype AI race car developed by Roborace which has both human and autonomous modes.
The DevBot has been a regular fixture at Formula E events over the past 12 months, showing off its driverless technology at races in Marrakech, Buenos Aires, Paris, Berlin and New York. But it had never gone head-to-head with a human driver.

Racing on the track that staged the opening two rounds of the 2017-18 Formula E championship, the task for both human and robot was complicated by the confines of the street circuit.

 

Top speed: 300 kph (186 mph)

“It’s quite scary because it’s very narrow — there was no room to maneuver,” Shields explained.
“Going into one of the chicanes is insanely tight, and of course if you get it wrong you are in the wall. You have to get it right every time.”
For Roborace, it wasn’t just the track but Hong Kong’s unique topography that posed problems.
“This was our first time of running our completely new AI Driver software in a built-up urban environment,” Roborace’s CTO Bryn Balcombe told CNN via email.
“The urban canyons created by the tall skyscrapers cause disruptions to GPS which are difficult to replicate in a testing environment.
Before completing a flying lap Roborace engineers deploy DevBot at low speed in what they call “explorer mode” where track data is gathered using state-of-the-art sensing equipment including LIDAR, radar, ultrasonics and cameras.
Once the track has been scanned, the information is processed by an AI computer (a NVIDIA Drive PX2) that uses complex algorithms to determine the fastest route (the racing line) around a track.
“Compared to a human, the main challenge is how fast it takes for the AI Driver to adapt and learn a new track,” Balcombe said.
In the end, it was a rare triumph for humans against AI opposition with Shields posting a fastest lap of 1:26.6 seconds — 7.4 seconds quicker than the DevBot could manage in AI mode around the 1.2-mile track.
For the victor there was pleasure and relief, but also a slight sense of disappointment.
“I’m a huge fan of this technology, so if the technology had won it would have been awesome,” Shields said.
“It was amazing to see DevBot go around in complete autonomous mode driving insanely fast around a ridiculously tight street circuit. It makes you think the possibilities are endless.”
DevBot may have lost this time around but its journey of discovery has only just begun as Roborace looks to push the boundaries in autonomous car technology even further.
“We are developing new competition formats that go beyond the limits imposed by human safety — challenges that will test the performance of intelligent vehicle technologies to the limit,” Balcombe said.

Eliminating human error

Convincing skeptical motorists to climb aboard with the technology is already one of the key challenges for automakers. Earlier this year, an AAA survey found that three-quarters of Americans said they would fear riding in an autonomous car.
Inevitably, incidents involving autonomous vehicles attract the headlines. One of Google’s self-driving cars crashed last year and a driverless public shuttle service in Las Vegas recently had an accident on its first day in service.
Roborace hasn’t been immune from crashes either. Last February while track testing in Buenos Aires DevBot plowed into the safety barriers.
But ceding control of your car to a computer may be safer and more efficient in the long run, says Professor Paul Jennings, from the UK’s Warwick University, one of the UK’s leading research centers for autonomous vehicles.
“Over 90% of accidents are down to human error so if in the long term we can eliminate human error we should save lives and reduce the number of accidents,” Jennings told CNN.
“We could also look at ways of better managing congestion. We could possibly improve energy efficiency. We can offer new services for people — potentially independent mobility to people who can’t have a driving license. There are lots of long-term benefits.”
Jennings is sympathetic to people’s concerns and he concedes that autonomous vehicles will need to be deployed gradually over the coming years.
“The early adoption and demonstration is likely to be in applications where you tie down the geography — I think that’s easier to manage,” he said.
“There may be adoption in enclosed environment, like a campus or maybe in the city in an area between a station and a city center. We’re even looking at trials of platooning in the fast lane of the motorway.”
“I can well understand people’s worries but I think it’s often you worry about lots of things that are new and only experience can help with that. I think it’s something you almost need to experience yourself.”
And that’s where Roborace is hoping to help too, introducing Formula E fans to its AI technology with DevBot and its futuristic “Robocar” designed by Daniel Simon.
“Ultimately the challenges are designed as an extreme R&D lab where innovations can be brought to future road cars, while the track events play a role in building public engagement and trust,” Balcombe said.