It’s never a good sign when car makers are called before Congress. It’s almost as bad as being asked to visit the President. But last week the meeting didn’t involve allegations or investigations. It was just an occasion for a friendly chat regarding “Self-Driving Cars: Road to Deployment.”
IEEE Spectrum was kind enough to excerpt notable moments from the Q&A which inadvertently highlighted the confusion prevailing among car makers as to the evolution of automated driving. Links to the full video and the IEEE Spectrum report appear below.
The hearing was held before the U.S. House Subcommittee on Digital Commerce and Consumer Protection and included Gill Pratt from Toyota Research Institute, General Motor’s Vice President of Global Strategy Mike Ableson, and Anders Karrberg, Vice President of Government Affairs at Volvo Car Group, along with Lyft’s Vice President of Public Policy Joseph Okpaku, and Nidhi Kalra, Co-Director and Senior Information Scientist, at the RAND Center for Decision Making Under Uncertainty.
Most notable among the responses elicited by the Congressional representatives were some key contradictions which conveyed the impression that the car makers are still seriously flummoxed by automated driving – and possibly less clear headed regarding the regulatory path forward than you might expect. Two issues, in particular, stand out: Level 3 vs. Level 4 development plans and the role of government.
Frank Pallone (D-N.J.) asked: “Volvo has said that it will skip Levl 3 automation and go from Level 2 to Level 4. Can you explain that decision?
Karrberg of Volvo replied: “At Level 3, the car is doing the driving. The car is doing the monitoring. But the driver is the fallback. So, you could end up in situations where the driver has to take back control, and that could happen within seconds. We are concerned about the Level 3 state, and therefore we are targeting Level 4.”
Kalra of RAND agreed: “There is evidence to suggest that Level 3 may show an increase in traffic crashes, and so it is defensible and plausible for automakers to skip Level 3. I don’t think there’s enough evidence to suggest that it be prohibited at this time, but it does post safety concerns that a lot of automakers are recognizing and trying to avoid.”
Reality is that Volvo has already launched its DriveMe project in Gothenburg, Sweden with drivers monitoring the automated driving in a Level 3 style:
http://www.volvocars.com/intl/about/our-innovation-brands/intellisafe/autonomous-driving/drive-me
Soon GM will be on a similar development path with human drivers monitoring the computer driving. Toyota, too, has adopted this vision. Level 3 automated driving is a critical step in the evolutionary path via which humans will “teach” the machines how to drive.
The greatest strength and weakness of Google/Alphabet/Waymo’s approach to automated driving has been the emphasis on driverlessness – no steering wheel, no pedals. This approach rules out the learning process and appears to be a limiting factor on Waymo’s commercial inroads thus far.
In fact, Pratt of Toyota, cast a bit of shade on Waymo in his testimony (without mentioning the company by name) by noting that the data sharing requirements for tracking driver interventions (in California) seems to favor companies taking a particular approach: “It’s the Federal government that we believe should take the leading role. As you may know, in California there’s a requirement, if you’re doing autonomous car development, that you report to the government what your disconnection rate is—every time that your car has a failure of a certain kind. That’s not such a bad idea, but that information then becomes publicly available, and it creates a perverse incentive, and the incentive is for companies to try to make that figure look good, because the public is watching. And that perverse incentive then causes the company to not try to test the difficult cases, but to test the easy cases, to make their score look good.”
But don’t get the idea that Pratt, or any car company, is prepared to fully embrace government intervention in automated driving development. Debbie Dingell (D-Mich) asked: “Are there specific things that Congress should avoid doing that would stifle the development of autonomous vehicles?”
Both GM and Toyota appear to agree that the government role should be limited. Says Ableson of GM: “We wouldn’t want to see [the] government taking steps to specify a specific technology or specific solution. I think as long as we keep in mind that the goal is to prove that the vehicles are safer than drivers today, the NHTSA guidelines published last year are a very good step in that direction, in that they specify what the expectations are before vehicles are deployed in a driverless fashion.”
Pratt of Toyota agreed: “An evidence based approach is really the best one, where the government sets what the criteria are for performance at the federal level, but does not dictate what the ways are to meet that particular level of performance.”
The message from Toyota and GM is clear. Tell us what to do, not how to do it.
These two perspectives are curious in that they conflict with both Toyota’s and GM’s position on the implementation of vehicle-to-vehicle communications currently facing a potential mandate from the National Highway Traffic Safety Administration. It ws no surprise, then, that Gus Bilirakis (R-Fla.) wanted to know where the respondents stood on V2V tech and how it “fit into the overall blueprint of deploying self-driving cars.”
Pratt took the lead, ultimately calling for preservation of the spectrum set aside for V2V applications: “Vehicle to vehicle as well as vehicle to infrastructure communication is of critical importance to autonomous vehicles. Of course, we drive using our own eyes to see other vehicles, but the potential is there for autonomous vehicles to use not only the sensors on the vehicle itself, but also sensors on neighboring vehicles in order to see the world better. And so, for example if you’re going around a corner, and there’s some trees or a building that’s blocking the view, vehicle to vehicle communication can give you the equivalent of x-ray vision, because you’re seeing not only your view, but also the view from other cars as well… We have to give ourselves every possible tool in the tool chest in order to try and solve this problem. So I think … saving spectrum for that use is also very important.”
In essence, Pratt distills a quite contentious and complex V2V debate into an argument for a government mandate for inter-vehicle communications. Such an argument is only consistent in the context of cellular-based solutions being considered as candidate alternative solutions for connecting cars to each other and infrastructure.
In the end, what emerges is a picture of ongoing confusion regarding the kind of help the automotive industry desires and the roll of Congress or even the U.S. Department of Transportation. It’s worth noting that the Federal Communications Commission is still testing the parameters governing spectrum sharing and an appropriate path forward. Testing of security protocols and standards for V2V are also ongoing.
The lack of a perspective or testimony from either Tesla Motors or Waymo is notable as is the lack of representation from state and local legislators and regulators and organizations advocating (both pro and con) on behalf of consumers – to say nothing of representatives of commercial trucking companies, rental car companies and taxi and limousine associations. In fact, the voices of consumer groups and employers are the ones most missed in this hearing.
If legislators could hear the demand side of the conversation more clearly it would clarify the confusion and contradictions currently ruling the space. Until these voices are heard, car makers will be left to their own aimless devices and regulators and legislators may well go awry creating delays and roadblocks or leading the industry down blind alleys.
Full video of the testimony can be found here: https://www.c-span.org/video/?423974-1/auto-industry-executives-testify-selfdriving-cars
Speculative Execution: Rethinking the Approach to CPU Scheduling