Human edges out robot car on race track

Thunderhill proved to be a good test of the car's control systems

A race between a robot car and a human has ended with a win for the humans - but only just.

The race was run on Thunderhill Raceway in California between an Audi TTS that can drive itself and a racing car driver familiar with the circuit.

The human driver completed a lap around the circuit a few seconds faster than the robotic car.

The race was part of research to develop control systems that will help to make domestic cars more autonomous.

Human race

The robot car in the race has been developed by researchers at the Centre for Automotive Research at Stanford University (Cars).

Called Shelley, the autonomous vehicle is fitted with sensors that work out its position on the road, feed back information about the grip of its tyres and help it plot the best route around the circuit.

Prof Chris Gerdes, head of the Cars Lab at Stanford, said Thunderhill was chosen because its 15 turns present the car's control systems with a wide variety of challenges. Some corners can be taken at high speed, some are chicanes, others are sharp and come at the end of long straights down which the car hit a top speed of 115mph (185kph).

Once familiar with the three-mile circuit the car was raced against one of Thunderhill's staff who was very familiar with the track and logged a slightly faster time.

"What human drivers do consistently well is feel out the limits of the car and push it just a little bit further and that is where they have an advantage," said Prof Gerdes.

He added that follow-up work had been done to record what the best human drivers did with the car's brakes, steering and throttle as they drove so this could be incorporated into the control systems the Stanford team is developing.

For instance, he said, in situations where the car is being driven at the limit of the grip of its tyres, the car cannot be turned via the steering wheel. Instead, said Prof Gerdes, race drivers use the brake and the throttle to force a car round a corner.

"We're learning what they are doing and it's those counter-intuitive behaviours that we are planning to put in the algorithm," he said.

"Our ultimate objective is not really to robotify [car racing] but to take these sorts of technologies, learning from the very best human drivers and turn those into safety systems that can work on cars," he told the Big Science Summit, a conference organised by The Atlantic magazine.

Currently, he said, driver assistance systems in vehicles actively prevent them performing manoeuvres that the best drivers use to avoid or get out of trouble.

Driving fast on a race track was one way to expose those high level abilities, he said. The maths of making a car steer safely at high speed around a tight bend was very similar to that needed to keep a car on the road if it hits a patch of ice. Both, he said, involved a calculation based on how much friction there was between the road and the tyres.

"As we set up these systems in the future, it's important not to build autonomous vehicles that are merely a collection of systems designed for human support but to think a little bit more holistically about making them as good as the very best human drivers," said Prof Gerdes. "It's not so much the technology as the capability of the human that is our inspiration now."