Are We There Yet?

The US Department of Transportation, National Highway Traffic Safety Administration, released a draft automated vehicle policy in September, 2016, and requested comments over a 60 day period.  My chief comment on this policy is that, for a policy that was more than a year in the making, the policy is rooted in the near term and does not look to the future enough. Specifically, the policy seems to refer to guidelines automation-optional cars, and fails to anticipate how driverless cars will profoundly change our culture.

The executive summary hails the potential advances in safety due to reduced collisions, increased mobility especially for persons with disabilities or conditions which prevent driving, and improved efficiency in roadways and parking from automated vehicle adoption.  The guide references the SAE levels of autonomy, and the draft policy addresses “highly automated vehicles (HAVs) from level 3, where the car can conduct some part of the driving but the human must be ready to re-take control, through level 4, where the human monitors car performance, through level 5, where the “automation can perform all driving tasks, under all conditions that a human driver could perform them.”  These words are a slight twist on the SAE standard.  Significantly, the major US automakers have decided that level 4 automation is essentially a tar baby, and all are planning to avoid it, going either to level 3 or level 5.  And the SAE standard itself says “full-time performance by an automated driving system of all aspects of the dynamic driving task under all roadway and environmental conditions that can be managed by a human driver.” [] There seems to be an implied difference in DoT’s paraphrasing that indicates that automation under level 5 is optional, whereas the SAE definition implies full control by automation.  This is the first indication that DoT is looking to regulate at level 3.

The NHTSA re-affirms the ways in which it regulates cars: it issues the Federal Motor Vehicle Safety Standards, which cars are manufactured to; it monitors the safety of vehicles on the road and conducts inquiries, occasionally issuing rules and recalls.  Automated cars falling under this proposed policy can be in test mode, when all the vehicles are driven by employees of the manufacturer, which echoes the methods Google and Otto practice; or deployed, in which automated cars can be driven by the public. The policy provides a framework for potential rules “Vehicle Performance Guidance,” in which categories of operation are called out. The policy is silent on the operation of cars that have no one in them. It concentrates on the operator, rather than the owner, which presumes there is an operator.

The categories of Vehicle Performance Guidance include “Scope and Process Guidance,” certification and registration; “Guidance Applicable to all HAV Systems on the Vehicle: Privacy, Human Machine Interface…” “Describe the Operational Design Domain: Day/Night, Speed, Weather..” “Object and Event Detection and Response,” and “Fall Back Minimal Risk Condition.”  The text accompanying explains the relevance of Operational Design Domain to level 1 and 2 vehicles.  This attention would be warranted if any major automaker were discussing production of L1 and L2 vehicles.  The guidance also discusses permanent placarding of the limits and capabilities of the automation in driver-readable spots in the vehicle, which sounds antiquated in an era in which new operating systems are routinely provided online to our laptops, phones, and to Tesla automobiles.framework-for-av-cars

The most backward-looking section however, concerns reversion to the “Fall Back Minimal Risk Condition.” The entire paragraph is cited here:

In situations where the HAV is outside of its defined ODD or in which conditions dynamically change to fall outside of the HAV’s ODD, the vehicle should transition to a minimal risk condition. The vehicle should give a clear indication of the type outlined in the HMI section to the occupants that it is switching to a minimal risk condition and that the HAV system is not available.

This paragraph makes sense when thinking about a Level 3 automobile, in which the human is on notice that the human may have to take control at any time.  This paragraph makes no sense however in a Level 5 automobile. If the minimal risk condition is total autonomous control, there is no difference between the normal control law and minimal risk.  What exactly is the car notifying the passengers of – passing control from automation to automation?

It is possible that the speed of government publishing is to blame for the apparent emphasis on levels of automation that are not relevant for 2020 and beyond.  After all, DoT’s reserved spectrum for Dedicated Short Range Communications (DSRC) has been held up for two years pending White House and OMB review.  DSRC is 75MHz of reserved spectrum in the 5.9 GHz band to enable vehicle to vehicle (V2V) and vehicle to infrastructure (V2I). For 15 years, automakers have been planning to incorporate DSRC in vehicles, but there is no rule establishing the DSRC spectrum and its use.  [] In the summer of 2016, Public Knowledge and New America Foundation Open Technology Institute filed requests to limit the use of the spectrum and require cyber security mechanisms. Many see this as attempts by the telecommunication – cell phone spectrum users specifically – to regain the use of that spectrum, which is prime G4 spectrum.  There is an argument that after 15 years of setting side this spectrum without use, the spectrum should revert to highest valued use.  In the 2017 model year vehicle Mercedes Benz introduced V2V, and Cadillac On-Star systems are capable of using the DSRC spectrum, so the delay is not technological or investment-based, but seems to lie in the speed of publishing policy rules.

One could ask whether there isn’t a lot of hype here. Are industry and commentators making a lot of noise about rushing a transition that consumers would not actually take up as rapidly as claimed? Some consumers don’t have a level of trust in automation needed to entrust their lives to a robotic car, and building such trust takes time.  It took six years for sales of the Prius in the U.S. to top 100,000 automobiles per year, or about 1.5% of new car sales per year.  Uber’s Otto service in Pittsburgh will roll out only 100 automated Volvos. Tesla is currently delivering about 50,000-60,000 vehicles per year, though this number comes from stockholder reports, and may represent worldwide deliveries. 2016 is the second calendar year of Tesla’s automated car sales.  av-adoption-ratesEven if the number of highly automated cars sold in the U.S. increased by 10% each year over the previous year, it would be 40 years before half the US car fleet was highly autonomous. On the other side, local concentrations can be higher; there is more interest and uptake of Teslas in New York, Washington DC, and San Francisco, and less interest in Michigan, the Dakotas, and other Midwestern states. The NHTSA is well aware its guidelines will have to apply to all areas and may be taking a conservative stance to balance out the building of trust as well as the early adopters.  The next three years will show which of the two – the auto manufacturers or the NHTSA – is being hopelessly optimistic.