Stanford University researchers have announced that they have developed a new way of controlling autonomous cars that integrates prior driving experiences, which will help cars perform more safely in extreme and unknown circumstances.

Two of Stanford’s autonomous vehicles, Niki and Shelley, were used to test the system at the limits of friction on a racetrack. Researchers say that the system performed about as well as an existing autonomous control system and an experienced racecar driver.

“Our work is motivated by safety, and we want autonomous vehicles to work in many scenarios, from normal driving on high-friction asphalt to fast, low-friction driving in ice and snow,” says Nathan Spielberg, a graduate student in mechanical engineering at Stanford and lead author of the paper about this research, which was recently published in Science Robotics.

“We want our algorithms to be as good as the best skilled drivers – and, hopefully, better.”

According to the Stanford researchers, current autonomous cars might rely on in-the-moment evaluations of their environment, but the control system that they have designed incorporates data from recent maneuvers and past driving experiences, including trips Niki took around an icy test track near the Arctic Circle.

The systems’ ability to learn from the past could prove to be very powerful, researchers say, especially taking into account the large amount of autonomous car data researchers are producing in the process of developing these vehicles.

The Stanford researchers explain that control systems for autonomous cars need access to information about the available road-tire friction, as this information determines the limits of how hard the car can brake, accelerate and steer in order to stay on the road in critical emergency scenarios. In order for autonomous cars to be pushed to their limits, researchers say, they have to be provided with certain details in advance, such as road-tire friction. This is hard in the real world, though, where friction is variable and often is difficult to predict.

Researchers built a neural network in an effort to develop a “more flexible, responsive control system.” A neural network is a type of “artificially intelligent computing system” that integrates data from past driving experiences at Thunderhill Raceway in Willows, California, and a winter test facility with foundational knowledge provided by 200,000 physics-based trajectories.

“With the techniques available today, you often have to choose between data-driven methods and approaches grounded in fundamental physics,” explains J. Christian Gerdes, professor of mechanical engineering and senior author of the paper.

“We think the path forward is to blend these approaches in order to harness their individual strengths. Physics can provide insight into structuring and validating neural network models that, in turn, can leverage massive amounts of data.”

The group conducted comparison tests for its new system at Thunderhill Raceway. First, Shelley sped around controlled by the physics-based autonomous system, pre-loaded with set information about the course and conditions.

Shelley and a skilled amateur driver generated comparable lap times when compared on the same course during 10 consecutive trials. Researchers then loaded Niki with their new neural network system, and the car performed similarly running both the learned and physics-based systems, despite the neural network lacking explicit information about road friction.

In simulated tests, the neural network system outperformed the physics-based system in both high-friction and low-friction scenarios, and the system did especially well in scenarios that mixed those two conditions.

While the results were encouraging, researchers stress that their neural network system does not perform well in conditions outside the ones it has experienced. But they note that as autonomous cars generate more data to train their network, they should be able to handle a broader range of conditions.

“With so many self-driving cars on the roads and in development, there is an abundance of data being generated from all kinds of driving scenarios,” Spielberg says.

“We wanted to build a neural network because there should be some way to make use of that data. If we can develop vehicles that have seen thousands of times more interactions than we have, we can hopefully make them safer.”