Skip to content Skip to main navigation Report an accessibility issue

Making Connected and Autonomous Vehicles Safe and Reliable

As connected and autonomous vehicles (CAVs) are quickly becoming reality, critical questions are being raised about how safe and reliable these vehicles are.

Many manufacturers are selling vehicles with partial or conditional (Level 2 and 3, respectively) automation and testing more advanced vehicles on public roads, but regulated standardized testing to ensure that the vehicles are safe is largely absent.

With funding from the Collaborative Sciences Center for Road Safety, Subhadeep Chakraborty, associate professor in the Department of Mechanical, Aerospace, and Biomedical Engineering, is working to change that.

Chakraborty has teamed up with UT’s Department of Civil and Environmental Engineering Professor Asad Khattak, and researchers from Duke University, and Oak Ridge National Laboratory to develop a comprehensive testing protocol that will standardize how to systematically and safely test Level 2 and 3 CAVs.

“Currently, there is no consensus about whether testing should exist at the state or federal level, what functions should be tested, how independent testing should occur, and what constitutes safe thresholds,” said Chakraborty. “With a testing protocol, automated vehicles can be systematically tested and certified to be generally safe before being put on the market or tested on public roads.”

The protocol will be designed to allow accelerated testing and identification of fringe cases and stress points where automated systems will be prone to failure.

“Essentially, we are trying to accelerate the pace at which CAVs can become more reliable and robust in their operations,” said Chakraborty.

Virtual reality (VR), a 3D simulated experience similar to the real world, will be used in the research along with physical vehicles in the loop. This setup will combine virtual and physical worlds, with the car mounted on a steerable dynamometer providing all the dynamics associated with real driving, while the virtual world simulates synthetic sensor data corresponding to various situational context.

“It is difficult to train machine learning for an unexpected combination of events that do not easily fall in a class that artificial intelligence can immediately recognize and classify,” said Chakraborty. “This problem is further aggravated by the fact that millions of miles need to be driven in all sorts of conditions before having the chance to encounter relatively rare events. With VR, we can develop these difficult driving situations for CAVs to negotiate, such as recognizing a pedestrian in dark clothing crossing a dark street.”

Using VR will also provide a large quantity of realistic synthetic data that cannot be obtained otherwise. Khattak will provide statistical analysis of the data that can result in valid conclusions and influence policy making in the future.

The team will spend the next two years testing and collecting data. When the project is complete, the results will be used to provide safety certification standard recommendations that regulatory agencies at the state and federal levels and the private sectors can use.

Chakraborty looks forward to seeing this research impact the immediate future and one day being instrumental in preventing accidents.