Driverless Car Update

The National Transportation Safety Board has issued its report about the 2016 crash that killed a Tesla driver. This has been billed as the “first self-driving car fatality,” but the truth is that the Tesla wasn’t designed to be a self-driving car. Instead, it is what is technically known as an SAE level 2 autonomous car, which is defined as “driver assistance systems of both steering and acceleration/ deceleration using information about the driving environment and with the expectation that the human driver perform all remaining aspects of the dynamic driving task.”

Instead of treating it this way, the driver acted as if it were a level 3 car, meaning a car capable of performing “all aspects of the dynamic driving task with the expectation that the human driver will respond appropriately to a request to intervene.” The Tesla was not designed to deal with all aspects of driving nor was it capable of making a request for the driver to intervene.

In this case, the car was going the legal speed limit on a highway and failed to slow or stop when a truck illegally entered the right of way to cross the highway. The Tesla was designed to detect another car in its lane but not a vehicle crossing the lane. The truck driver–who, the NTSB notes, had been smoking marijuana–cross the highway in violation of the Tesla’s right of way. An alert driver would have slowed down, but the Tesla driver was relying on his car to do things it wasn’t designed to do.

While the truck driver was clearly at fault for violating the right of way, NTSB notes that the “Tesla driver’s pattern of use of the Autopilot system indicates an overreliance on the automation and a lack of understanding of system limitations.” Tesla itself was also guilty, the NTSB notes: “If automated vehicle control systems do not automatically restrict their own operation to those conditions for which they were designed and are appropriate, the risk of driver misuse remains.”

Most cars with level 2 automation, including Audi, BMW, Infiniti, Mercedes, and Volvo, require drivers to place their hands on the wheel at least every 15 seconds. If the car detects that the driver’s hands are off the wheel for more than 15 seconds, the car shuts off the self-steering system. Tesla did not include this safety limit.

The NTSB isn’t satisfied with even that limit, however. “Because driving is an inherently visual task and a driver may touch the steering wheel without visually assessing the roadway, traffic conditions, or vehicle control system performance,” said the report, “monitoring steering wheel torque provides a poor surrogate means of determining the automated vehicle driver’s degree of engagement with the driving task.” The report recommended that manufacturers “Develop applications to more effectively sense the driver’s level of engagement and alert the driver when engagement is lacking.”

Cadillac has already developed such a system. The company’s Super Cruise system has a camera that watches the driver and makes sure they are watching the road even when the car itself is doing the steering. If the driver fails to monitor the road and doesn’t respond to warnings, the car will slow and eventually come to a full stop. The Cadillac is still a level 2 car, but it verges on a level 3 system.

In any case, level 2 systems are far from true driverless cars. However, General Motors subsidiary Cruise also claims to have developed the world’s first mass-producible driverless car. This car is “assembled in a high-volume assembly plant capable of producing 100,000’s of vehicles per year.” Cruise doesn’t plan to make 100,000 cars anytime soon, but it will make some to use for its San Francisco ride-sharing program.

Cruise’s car, like the ones being designed by Ford, Volkswagen, and other manufacturers, depends on high-definition maps. The cars will only work in driverless mode in areas that have been mapped, so until most of the streets and highways in the country have been mapped, the market for such cars will be small. That’s why both Ford and GM plan to introduce these cars in ride-sharing service rather than for sale.

The process of mapping might be speeded up if all manufacturers worked together to create one giant map. That may be what BMW, Mercedes, and Volkswagen had in mind when they bought Here from Nokia. But Cruise is making its own maps, and Ford invested in a company called Civil Maps. In the end, having competing map systems will probably be a good thing, partly because competition will encourage them to quickly map all 4 million miles of roads in the United States rather than be satisfied with, say, just the roads in major urban areas.

As of September 1, a total of 40 companies had received licenses to test autonomous cars in California. But a company called Comet Labs has tracked a total of 263 companies that are working on driverless car technologies. Most of these companies are just developing one aspect of those technologies, such as sensors, maps, or software. But it is clear that billions of dollars a year are being spent, making the prospects of level 5 driverless cars–“the full-time performance by an automated driving system of all aspects of the dynamic driving task under all roadway and environmental conditions”–all the closer.

Share

7 thoughts on “Driverless Car Update

  1. Dave Brough

    AP: “.. the car was going the legal speed limit on a highway and failed to slow or stop when a truck illegally entered the right of way to cross the highway.”
    In fact, the driver was doing 74 in a 65 mph zone.
    http://www.reuters.com/article/us-tesla-autopilot/u-s-report-says-tesla-in-fatal-autopilot-crash-was-speeding-idUSKCN1062CT
    Another factor. There is a slight hill preceding the intersection that could have played a factor from the semi driver’s point of view. I suggest that when the driver checked his sight line prior to his turn, the Tesla was hidden from view, and once committed, had two choices, continue or stop, neither of which would not have been in the best interests of a car to dumb to stop. I say that in this case, the intersection belonged to the truck.
    https://www.google.com/maps/@29.4108082,-82.539351,3a,37.5y,266.47h,90.26t/data=!3m6!1e1!3m4!1sHwUywO4x9xs_O-BNhg01bQ!2e0!7i13312!8i6656
    I also say the blame rests entirely on Tesla, which should have realized that when you dangle a carrot in front of a donkey, the donkey is likely to try to take the bite.

  2. CapitalistRoader

    Tesla foolishly allowed their Autopilot to be set above the speed limit on a non-divided highway. I think that’s no longer the case. I too blame Tesla for that single death. Similarly, faulty meat servos were almost certainly the cause of the other 40,199 traffic deaths last year.

    Level 2 autonomy is fine as an interim step. Level 3 probably should never be put in wide production beacause car manufacturers would be looking at non-stop lawsuits. Go right to level 4.

  3. rallenr

    Check out our FB page on Level 4 ‘driverless’ trucks: fb.me/level4transport. A number of senior executives, like John Krafcik at Google/Waymo are predicting we’ll see autonomous semis in regular, commercial use before driverless automobiles.

    Roy Reynolds
    level4transport.com

  4. prk166


    In the end, having competing map systems will probably be a good thing, partly because competition will encourage them to quickly map all 4 million miles of roads in the United States rather than be satisfied with, say, just the roads in major urban areas.
    ” ~anti-planner

    Technically those roads are all mapped. The question is what kind of data is needed for these driverless cars. People are gathering that up ( below ), not just employees of companies. Data is both automatically detected and manually added. I’ve updated Open Street Maps ( OSM ) to remove roads or update the route for a bike path. Others literally go onto OSM and add data for where a wheelchair lift is located on a building and the extent of street lighting.

    https://www.mapillary.com/app/?lat=34.70858821803262&lng=-84.14985812658512&z=17&pKey=gPXYdowVi4YWlH9Lg3vzPg&focus=photo&x=0.4888728194616377&y=0.6268760025207083&zoom=0

    https://www.openstreetmap.org/edit?editor=id#map=18/34.70858821803262/-84.14985812658512

    I’m not sure how much an issue the data will turn out to be. It’s not an easy problem but there are solutions in place today. What data is missing could be addressed by someone like Google, Verizon, Here, Ford, GoGoMaps or whomever getting cameras into cars with a system that uses cell coverage to auto upload and store for later when coverage doesn’t exist. The technology is becoming cheap and pretty good. Mix in AI that auto detects street signs, etc and most places in the next decade could have fresh data.

    And of course with the processing power out there now, let alone what will likely be around in a decade, that same AI that’s detecting signs in a database of pictures can be tweaked and ran in real-time on any signs the car is seeing.

    Sooooo…. I’m curious as to what sort of mapping is really needed for fully autonomous cars. I suspect a large part of talk about it is just the need for data, for real world examples to train the AI, not necessarily needed directly for day-in, day-out driving.

  5. JOHN1000

    Tesla survives on government subsidies and is the liberal, left-wing media darling. So no headlines about their being at fault.

    Is it then any surprise that Tesla was the one company that ignored the safety feature described in the article. They knew they could get away with it – and that a Ford or GM would be pilloried for doing the same things.

    Tesla = fraud.

  6. prk166


    A number of senior executives, like John Krafcik at Google/Waymo are predicting we’ll see autonomous semis in regular, commercial use before driverless automobiles.
    “~rallenr

    Thank you, Mr. Reynolds. That would make a lot of sense to me. A driverless semi eliminates the calendar time restrictions they have. With a driver, they can only driver so X hours / day, etc. No driver, you can just keep on truckin’ non-stop!

    I’m curious how some of this automation could affect railroads. I’m re-reading Marc Levinson’s The Box on the shipping container. I’d be curious to see how automation encourages more containerization in the shipping and logistics industry.

    BTW, Mr. o’Toole you may enjoy reading the history with an eye on the things that government planning can muck up. It looks like we were a hair away from some folks trying to enforce some bad standards, including that the same containers that are piled onto those huge ships, also be designed to be transported in an airplane. DOH!

Leave a Reply