A self-driving car has learned to make high-speed turns without spinning out. The skill could come in useful during emergency manoeuvres.
J Christian Gerdes and colleagues at Stanford University used a type of artificial intelligence algorithm called a neural network, which is loosely based on the neural networks in our brains, to create the self-driving system.
They trained the neural network on data from more than 200,000 motion samples taken from test drives on a variety of surfaces, including on a mix of snow and ice at a track near the Arctic Circle.
Advertisement
The team equipped a Volkswagen GTI with the algorithm and tested it on an oval-shaped race track.
The car drove at speeds as fast as physically possible for the road surface, turning circle, and tyres, observing its motion from previous fractions of a second to adjust its steering and acceleration.
At turning speeds around 50 kilometres per hour it had a low tracking error, deviating less than 50cm from its goal turning path.
The team found their neural network also worked when the track was covered in snow or ice. Accurate motion predictions on different road surfaces are important to ensure the car performs well in a variety of weather conditions.
For autonomous vehicles to operate safely, they need control systems that can rapidly brake, accelerate or steer in critical situations, letting them drive safely at the limits of friction – just before the tyres stop gripping the road and the car spins out. Gerdes and his team’s system could help in emergency situations, where sudden swerves are required.
One challenge of the neural network is a lack of insight into how it works, says Gerdes. “If you give it a set of conditions it hasn’t seen before, it may extrapolate in ways that are completely wrong,” resulting in potentially dangerous steering controls, he says.
The team are now building safety features into the system to check its decisions are reasonable.
Journal reference: Science Robotics, DOI: 10.1126/scirobotics.aaw8703
More on these topics: