A new test could significantly streamline the process required for self-driving cars to be deemed road-ready.
Researchers from the University of Michigan have developed a new test—which uses data from more than 25 million miles of real-world driving— that is expected to cut the time needed to evaluate robotic vehicles’ handling of potentially dangerous situations.
“Even the most advanced and largest-scale efforts to test automated vehicles today fall woefully short of what is needed to thoroughly test these robotic cars,” Huei Peng, director of Mcity and the Roger L. McCarthy Professor of Mechanical Engineering at Michigan, said in a statement.
The researchers keyed in on the interaction between an automated car following a human driver and a human driver merging in front of an automated car—the two most common situations that are expected to result in serious crashes.
The vehicles were exposed to a condensed set of the most challenging driving situations, where just 1,000 miles of testing can yield the equivalent of 300,000 to 100 million miles of real-world driving.
Autonomous vehicles require extensive testing so they can zero in on rare difficult scenarios. According to the researchers, tests will need to prove that the autonomous vehicle with 80 percent confidence that they’re 90 percent safer than human drivers for consumers to accept driverless vehicles.
To achieve this confidence, the vehicles will need to be driven in simulated or real-world settings for 11 billion miles. However, it would take almost 10 years of round-the-clock testing to reach two million miles in urban conditions.
The new accelerated evaluation process breaks down difficult real-world driving situations into components that can be tested or simulated repeatedly.
The four-step accelerated approach identifies events that could contain meaningful interactions between an automated vehicle and one driven by a human. These simulation conducted mathematical tests to assess the risk and probability of certain outcomes and interpreted the accelerated test results using a technique called importance sampling to learn how the vehicle would perform statistically in everyday situations.