How Michigan’s State-Of-The-Art Facility Lets You Test Real Autonomous Car Remotely Using Both Real And Virtual Obstacles

Just over nine years ago, a first-of-its-kind test track opened in northeast Ann Arbor, Michigan. Several years after pharmaceutical company Pfizer decided to shut down its Ann Arbor R&D facility, the University of Michigan took over most of the property for a new research and engineering campus. A portion of this area is home to the U of M Transportation Research Institute (UMTRI) and MCity. Today, UMTRI unveiled MCity 2.0, a major upgrade to the facility that adds an array of new capabilities.MCity was conceived as the world’s first purpose-built test facility specifically for connected and automated vehicles. The 32-acre site features a variety of roads, intersections, tunnels, and traffic signals as well as connectivity infrastructure and building facades. Since its opening, it has been used not only by university researchers but also by engineers from Ford, Honda, Toyota, Nissan, and a variety of suppliers.

Until now, all that work took place on-site in Ann Arbor. That changes with MCity 2.0. With funding from the National Science Foundation, MCity has added a range of new capabilities that make it available to researchers all over the country via a new digital infrastructure that enables remote testing and simulation as well as mixing of the two.

Much of what has been deployed is running on Amazon Web Services including cloud-based simulation as a service and a unique mixed reality environment. At the track, cars are connected by a Verizon 5G connection that is claimed to have only about 9 milliseconds of latency. [Ed Note: For comparison, the blink of an eye is 100-150 milliseconds. – Pete]

Remote researchers anywhere in the U.S. can access a vehicle that is located at MCity, download their software algorithms, and begin running tests. Real-time sensor and telemetry data is beamed back to remote testers so that they can see what is happening while at the same time being ingested into a cloud-based data engine that is isolated for each user.

Until now, the testing on the track only involved whatever user agents (other vehicles, pedestrians, cyclists, etc.) were physically present on the track. However, MCity’s new mixed reality environment can now blend in virtual agents generated by a VI-Grade Worldsim system into that real-world environment. We saw a demonstration of a physical automated vehicle driving on the track and interacting with virtual vehicles being injected over the wireless connection.

During the demonstration, the actual vehicle drove around some virtual vehicles, made a left turn, then came to a stop behind a stalled virtual truck. At that point, a teleoperating driver took over control to manually maneuver the AV around the virtual truck and then returned control to the autonomous system.

Full simulation environments are and will continue to be a key element of developing and validating the safety of automated driving systems and provide the ability to repeatedly test the same scenario as iterations are made on the software. It also enables testing of a nearly infinite variety of scenarios. But testing of actual autonomous vehicles is also crucial, and being able to test them in complex environments using virtual agents is a far safer and more cost-effective strategy compared to adding other physical vehicles to the test track. Another benefit is the ability to precisely repeat interactions between the virtual agents and autonomous vehicles, which would be very difficult if not impossible to do with actual cars, pedestrians, cyclists, and other road users.

The updates for MCity 2.0 are a major advancement for researchers testing automated driving systems and we’ll probably see other facilities like California’s Gomentum Station and the American Center for Mobility move to add similar capabilities in the near future. However, for academic researchers, MCity will likely remain the preferred testing site.