Engineers in Canada have developed decision-making and motion-planning technology to limit injuries and damage when autonomous vehicles are involved in accidents with pedestrians.
After recognising that a collision is inevitable, the model predictive control (MPC) algorithm is said to analyse all available options and chooses the course of action with the least serious outcome.
“What can we do in order to minimise the consequences?” said Amir Khajepour, a professor of mechanical and mechatronics engineering at the University of Waterloo. “That is our focus.”
The first rule for the autonomous vehicle (AV) crash-mitigation technology is avoiding collisions with pedestrians. It then considers factors such as relative speeds, angles of collision and differences in mass and vehicle type to determine the best possible manoeuvre, such as braking or steering in one direction or another.
“We consider the whole traffic environment perceived by the autonomous vehicle, including all the other vehicles and obstacles around it,” said Dongpu Cao, a mechanical and mechatronics engineering professor at Waterloo.
Khajepour, director of the Mechatronic Vehicle Systems Lab, said the system is needed because the idea that AVs will completely eliminate crashes is wrong.
Although safety should improve dramatically, he said, there are too many uncertainties for self-driving vehicles to handle them all without some mishaps.
“There are hundreds, thousands, of variables we have no control over,” he said in a statement. “We are driving and all of a sudden there is black ice, for instance, or a boulder rolls down a mountain onto the road.”
According to Waterloo, AVs can limit damage when a crash is unavoidable because they always know what is happening around them via sensors, cameras and other sources, and routinely make tens and even hundreds of decisions per second based on that information.
The new system decides how an AV should respond in emergency situations based primarily on pre-defined mathematical calculations considering the severity of crash injuries and damage.
The researchers did not, however, attempt to factor in complex ethical questions, such as whether an AV should put the safety of its own occupants first or weigh the well-being of all people in a crash equally. Khajepour said the system framework is designed to integrated into AVs once carmakers and regulators resolve ethical rules for self-driving vehicles.
A paper on their work, Crash Mitigation in Motion Planning for Autonomous Vehicles, has been published in IEEE Transactions on Intelligent Transportation Systems.