How should police handle AV incidents?

As more autonomous vehicles operate in pilot programs and limited commercial deployments in multiple states, questions linger over how they should interact with law enforcement.

Some autonomous driving tech companies have encountered problems.

In April, a robotaxi operated by General Motors-backed Cruise was pulled over by San Francisco Police Department officers for driving without its headlights on at night. The vehicle repositioned itself before the traffic stop concluded, driving across an intersection before pulling over again.

There have also been instances in which Cruise vehicles blocked emergency responders and intersections in the city.

Self-driving tech companies say they are trying to work out these kinks and prepare for smooth interactions once their fleets are widely deployed.

Autonomous truck developer Embark Trucks Inc. has partnered with the Texas Department of Public Safety and the Travis County Sheriff's Office in the state to test how its vehicles respond to law enforcement.

Cameras and lidar sensors on Embark's trucks can identify emergency vehicle lights and respond accordingly, said Emily Warren, the company's head of public policy.

When it detects the lights, the vehicle pulls to the side of the road and contacts the remote Embark Control Center. Warren said Embark prints a phone number for officers on the cabs of its trucks. The number connects directly with the control center. The truck's lockbox holds compliance documents.

Embark wanted "to design an intuitive protocol that easily works with law enforcement and does not require new training," Warren said.

Working with law enforcement, Embark learned officers want to approach trucks from the passenger side, as they do during human-driven big-rig stops.

Warren said Embark had collaborated with other law enforcement agencies, such as the California Highway Patrol, to get input on officers' expectations when interacting with AVs.

The Travis County Sheriff's Office was glad to work with Embark, spokesperson Kristen Dark said.

Aurora Innovation Inc. also works on communication between its driving system and law enforcement. It is developing a software program called Aurora Beacon, which will guide the vehicle in certain instances, such as when law enforcement needs to interact with it, the company said in a blog post. The software will trigger a session in which support personnel can evaluate the issue.

"Any of a variety of triggers can initiate a support session, from a truck health alert to an unusual sighting in the road to the truck's proximity to a point of interest," the company said. "In these sessions, support personnel will see what the truck sees, hear what the truck hears, and provide high-level guidance for the vehicle's objective or course of action."

Aurora also said it has held training sessions for law enforcement officers and developed guides on how to interact with an AV. The training included taking officers on rides and developing safety case frameworks.

Waymo, one of Aurora's prime competitors, has conducted training for first responders in multiple states.

"We have a team with decades of law enforcement experience that has provided training to hundreds of officers and firefighters in California, Arizona and Texas detailing best practices for safe interactions with Waymo vehicles," said Waymo spokesperson Katherine Barna. Waymo, the self-driving division of Google parent company Alphabet, is developing autonomous trucks and robotaxis.

Training for the Phoenix Police Department consisted of learning how to shut off Waymo vehicles properly and how to access specific police resources within the car, Sgt. Phil Krynsky told Automotive News. Waymo is testing robotaxis in the city.

There are still complications that need solving before AVs can be implemented at scale, said Ram Vasudevan, a mechanical engineering and robotics professor at the University of Michigan in Ann Arbor.

Initial operations would be focused on using the vehicles for fleet businesses, where companies could calculate a return on their investment, he said.

But "even that existing, at scale and done well in a place like Ann Arbor is probably really far off," Vasudevan said.