in-vehicle computer in Nevada was granted to a robotic Prius, so why is a Lexus SUV wearing the plates? It’s all legal, and that might be a problem
In fact, an investigation by IEEE Spectrum uncovered that none of the Priuses that Nevada originally licensed as AU-001, AU-002, or AU-003 were the vehicle evaluated by DMV officials in 2012. This means that none of Google’s self-driving vehicles licensed to drive on Nevada’s roads have actually taken the state’s self-driving test.
Google is not breaking the law. While Nevada’s self-driving test covers many of the same scenarios as in a human exam, such as city driving, highway driving, crosswalks, traffic lights, and roundabouts, it was designed to evaluate the vehicle pc underlying artificial intelligence of autonomous driving rather than specific vehicles, hardware, or versions of software. Thus, once a single Google car had passed the test, the company was free to register other vehicles for its own trials. Google did this again when it renewed its testing license in 2014, transferring the nation’s first “AU” license plates to three Lexus hybrids packed with new or upgraded sensors and software.
Of the few states that have welcomed experimental self-driving vehicles, only Nevada requires a vehicle pc test drive, and there is no suggestion that the Lexus SUVs pose any greater risk to the public than the Priuses. Nevertheless, this casual substitution of complex systems has some experts concerned. Bryant Walker Smith is a law professor at the University of South Carolina and chair of the Emerging Technology Law Committee of the Transportation Research Board of the National Academies. He says, “Autonomous vehicles are necessarily a combination of hardware and software. You couldn’t simply take Google’s algorithms for the Prius and apply them to the Lexus SUV. Anything down to the tire pressure can be relevant for how a vehicle will respond in emergency situations. Braking force, the condition of the brakes, and sightlines are all functions of the hardware and can potentially vary from vehicle to in-vehicle computer, even within the same make, model, and year.”
“It shows the disconnect between Google’s thinking about driverless cars and everyone else's,” says Ryan Calo, a law professor at the University of Washington who specializes in robotics and public policy. “Google's engineers are thinking, ‘When we model the world, how well does our vehicle respond? The in-vehicle computer physical shell that lives in is less important. What ultimately matters is the quality of that software.’ ”
Google was the driving force behind the Nevada regulations. “The whole set of developments in Nevada have been at the behest of, and working closely with, Google,” says Calo. And there are some very good reasons to allow flexibility in the testing and licensing of autonomous vehicles, especially experimental ones. The software in today’s self-driving vehicles is typically changed frequently, even daily. No one would want a critical safety update, for example, to be delayed by a complex regulatory process. And yet the wholesale grandfathering in of new vehicles, technologies, sensors, and software raises concerns over what exactly is being tested and why.
For its part, Nevada insists that safety is the most important part of its autonomous vehicle testing program. “At this time, the department does not view the changes as justification for Google to provide another demonstration,” says Jude Hurin, the DMV manager who oversees experimental autonomous vehicles in the state.
But that doesn't mean Nevada isn't keeping an eye on things. “Google recently reported that they would be testing an autonomous vehicle that has no steering wheel. My opinion is that Nevada would not allow testing of this vehicle without a steering wheel since it does not meet the intent of our existing safety requirements,” says Hurin.
However, given that the license-renewal process does not currently require Google to submit any technical data for new in-vehicle system in cars, it is unclear how Nevada would identify the vehicles it wanted to recertify in the first place.
“The traditional regulatory model simply isn’t prepared to address this technology,” says Smith. “One thing we might see is more states, and even the federal government, moving to embrace process standards. That is, looking not at how something performs but what was the thought that went into it; the processes used to design in-vehicle system, test, and verify it; and what safety protocols were implemented. Realistically, these are the only things that can be well measured.”Until then, Google’s historic AU-001 self-driving car can keep on transforming—and keep driving on Nevada's roads.