Control Centers Could Aid Transition to Driverless Cars

Researchers at the University of Michigan say remotely located technicians could help troubleshoot and respond to potential problems for self-driving cars in real time.

Their work, which is partially funded by the U.S. Dept. of Transportation, uses various sensors and connected vehicle technology to monitor the upcoming driving environment. Artificial intelligence software analyzes the data and can predict complex events 10-30 seconds in advance that are likely to cause an autonomous driving system to disengage and return control to a human driver.

The information is sent to a remotely located control center that immediately generates several possible scenarios that the host vehicle could encounter. Using driving simulators, technicians at the center respond to the simulations and send the results back to the vehicle.

This would build a library of human-generated responses that a vehicle operating in autonomous mode could choose from, based on real-time sensor data. The researchers say the process could be completed in time to keep vehicles safely operating without requiring human intervention.

Staffing a control center with a handful of technicians would be far more cost effective than putting a human backup in every vehicle, the researchers note. They say such systems will help speed development and improve safety as the industry refines autonomous vehicle technology to handle more complex situations.

The best self-driving cars being tested in the U.S. currently average one disengagement about every 5,000 miles. At that rate, the researchers estimate, it would require about 100,000 people to staff control centers if every vehicle in the U.S. was autonomous.

A software platform for the program is under development. Initial off-board tests are expected to begin later this year, followed by in-vehicle tests in 2020.