Autonomous Vehicle Safety Tested on Machine-Learning Model Roadway

University of Michigan researchers have developed a machine-learning model roadway, designed to test the safety of autonomous vehicles. 

Hailed as the “first statistically realistic roadway simulation,” the model is designed to simulate one of Michigan’s most crash-prone intersections, located in Ann Arbor. 

Data collected from a roundabout at the intersection was used to train the simulation, known as the Neural Naturalistic Driving Environment (NeuralNDE), with sensor systems installed on light poles at the roundabout.

By using this data as the backbone of the virtual roadway, the team was able to realistically simulate a driver’s experience.

“The NeuralNDE reproduces the driving environment and, more importantly, realistically simulates these safety-critical situations so we can evaluate the safety performance of autonomous vehicles,” said Henry Liu, University of Michigan professor of civil engineering.

“The reason that we chose that location is that roundabouts are a very challenging, urban driving scenario for autonomous vehicles,” said Xintao Yan, first author of a study on the project. “In a roundabout, drivers are required to spontaneously negotiate and cooperate with other drivers moving through the intersection. In addition, this particular roundabout experiences high traffic volume and has two lanes, which adds to its complexity.”

The NeuralNDE is also harnessed for the team’s CCAT Safe AI Framework for Trustworthy Edge Scenario Tests, or SAFE TEST. The system uses AI to reduce the testing time for autonomous vehicles, simulating vast numbers of safety-critical incidents to test and train the AV software more rapidly.

The team is also using NeuralNDE for a project testing AV software using mixed reality. Based at the Mcity Test Facility, autonomous vehicles on the track have to navigate not only real conditions but also virtual obstacles and hazards, including other drivers, cyclists and pedestrians.