Slashdot Mirror

← Back to Stories (view on slashdot.org)

Toyota Is Uneasy About the Handoff Between Automated Systems and Drivers (caranddriver.com)

Posted by BeauHD on from the safety-first dept.

schwit1 shares a report from Car and Driver: Toyota has not yet decided whether it will bring a car to market that is capable of automated driving in some situations yet still requires a human driver behind a wheel who can take control if needed -- but the automaker, characteristically, is more cautious than many about moving forward with the technology. Citing safety concerns regarding the handoff between self-driving technology and human driver, Kiyotaka Ise, Toyota's chief safety technology officer, said the biggest issue with these kinds of systems is that "there is a limbo for several seconds between machine and human" in incidents when a car prompts a human to retake control if it cannot handle operations. These kinds of systems, defined as Level 3 autonomy by SAE, have divided automakers and tech companies in their approaches to developing cars for the self-driving future. As opposed to Level 2 systems, like Tesla Motors' Autopilot, in which a human driver is expected to keep his or her eyes and attention on the road while a system conducts most aspects of the driving, Level 3 is characterized by the system's claiming responsibility for the driving task when it is enabled. Although Toyota assures us that its researchers are hard at work figuring out the challenges of Level 3 autonomy, it seems like the company could eventually join others moving directly from its current Level 2 system to a Level 4 system. Given the self-driving race has been on for a while, this could put Toyota at a competitive disadvantage, but it's clear engineers at the company care more about getting things right than they do about being first.

18 of 135 comments (clear)

  1. Toyota is... by Anonymous Coward · · Score: 5, Insightful

    The summary states, "it's clear engineers at the company care more about getting things right than they do about being first."

    So, basically what you're saying is, Toyota is the anti-Tesla.

    1. Re:Toyota is... by AmiMoJo · · Score: 3, Insightful

      I'm not convinced that Tesla will get to level 5 with their current hardware. They are already selling level 5 to customers as a future firmware update (â3000 extra last time I looked), saying they will upgrade hardware if necessary for people who already paid.

      Their system is only cameras and ultrasonic sensors, no lidar. They are using neural nets for image processing.

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    2. Re:Toyota is... by Anonymous Coward · · Score: 5, Funny

      Tesla is the industry leader in making promises .

    3. Re:Toyota is... by mjwx · · Score: 4, Insightful

      The summary states, "it's clear engineers at the company care more about getting things right than they do about being first."

      So, basically what you're saying is, Toyota is the anti-Tesla.

      Basically you're saying Toyota is being Toyota (conservative to the extreme, but good at what they do).

      Toyota is not the only one concerned with this, as a road user I'm concerned what will happen when Dopey Doris' automated car struggles with faded lines on a single lane road (quite common on my 18 mile commute to work). Right now, Dopey Doris can only spend half her attention on her phone, I hate to think what will happen when she puts her full attention into it and because she's so engrossed in FaceCrush or watching the latest episode of Keeping up the Cardasians that she completely tunes out the alarm throwing control back to the driver and the car veers into my lane uncontrolled.

      Autonomous cars need to be 99.999999999999999% reliable before they should be considered ready for public consumption. Right now, they're nowhere near it. Google's success has been due to two factors, 1. it was all done in sunny California (I'd like to see the same car in Berkshire) and; 2. the car has been in the hands of a professional driver the whole time. The current track record for autonomous cars is nil, the record is for car and driver working together. Of course we know in the real world if you gave the Google autonomous car to Dopey Doris commuting from Finchamstead every day, she's going to assume that it will do everything for her, so we need to make sure it can operate without human intervention because it needs to, human intervention cant be counted upon from the average steering wheel attendant with a phone shoved up their nose (we get enough collisions from these types as it is without giving them a reassurance).

      --
      Calling someone a "hater" only means you can not rationally rebut their argument.
    4. Re:Toyota is... by rmdingler · · Score: 2

      Autonomous cars need to be 99.999999999999999% reliable before they should be considered ready for public consumption.

      Since that's essentially an impossible standard, what you're saying is that we should never use autonomous land vehicles, even though human drivers fall massively short of that same safety threshold.

      --
      Happiness in intelligent people is the rarest thing I know.

      Ernest Hemingway

    5. Re:Toyota is... by PolygamousRanchKid+ · · Score: 4, Interesting

      The Duke of Wellington claimed he won the Battle of Waterloo against Napoleon on "the playing fields of Eton".

      The big battle for autonomous driving will be won or lost in the tort courts of the US. Who is responsible for the accident? The driver? Or the manufacturer?

      Your local ambulance chaser lawyer would prefer to sue the manufacturer . . . simply because the manufacturer has more money!

      The first big cases will unsettle the industry, but a sort of fudge agreement will be reached between lawyer groups, the manufacturers and the insurance companies. Unfortunately, the average driver will end up paying for this.

      The lawyers don't want to kill the autonomous car industry . . . they want to "milk" it for their "piece of the action".

      --
      Schroedinger's Brexit: The UK is both in and out of the EU at the same time!
    6. Re:Toyota is... by CastrTroy · · Score: 4, Insightful

      This is also a big concern of mine. Cars should either be 100% autonomous, or 0% autonomous. I'm all for adaptive cruise control, but as soon as you introduce technology that allows people to take their attention away from the driving and have it still follow the road for a significant period of time, that's where you run into problems.

      If you haven't had to actually touch the steering wheel for a month, how much would you really be paying attention? What happens when the car screws up and you need to take over? Are you going to be too engrossed if your other activities to take over? Also, what is the point of paying for all this technology anyway if you don't get to actually not pay attention anyway. If you're going to have to keep your hands on the wheel, you might as well actually be driving, because other wise it isn't really worth the expense.

      --

      Anthropic principle: We see the universe the way it is because if it were different we would not be here to see it.
    7. Re:Toyota is... by AmiMoJo · · Score: 2

      The whole sensor package seems poorly thought out. For example, the lack of a nose camera means that they can't implement a 360 degree overhead view like Nissan and several other manufacturers have now.

      And then there is the whole "AP2.5" debacle, where they did a major computing power upgrade over the original AP2.0 hardware after realizing it wasn't up to the task, and then promised free upgrades to people who had already paid for full self driving a year ago.

      The thing about cameras is that they are not the same as human eyes. People have been extracting images from the Tesla cameras and the resolution just isn't good enough to resolve signs at distance, for example. Also processing those images is proving to be difficult. Currently they are down-sampling them to an even lower resolution before handing them to their neural networks for recognition.

      The current state of their system is that they can follow well marked roads somewhat reliably. On corners the car will sometimes stray into oncoming traffic or into walls. Junctions confuse it. Even on well marked roads in good weather it sometimes ping-pongs left and right in the lane. And they use GPS maps for assistance, something that a full self driving car can't rely on.

      Musk said they would demo driving from coast to coast autonomously this year. Seems unlikely to happen, although it might just be possible if they stick to major roads. But they are so far from being able to handle anything tricky, like roundabouts or unusual junctions or car parks or even read basic road signs and traffic lights that I'm thinking some of the people who already paid for the self-driving feature will be selling the car long before it ships.

      --
      const int one = 65536; (Silvermoon, Texture.cs)
      SJW, n: "Someone I don't like, and by the way I'm a fuckwit" - AC
    8. Re:Toyota is... by Rei · · Score: 2

      12 ultrasonic sensors on Model 3.

      That said, what we have that a car doesn't is a brain that automatically fixes photogrammetric stitching errors based on the logic of "that doesn't make sense" - whether something "makes sense" or not being an AI-hard problem. So one tries to compensate for this on vehicles with better sensors than human beings have.

      Lidar provides a superb data stream when conditions are right, but it's too problematic. You're not going to put awkward, draggy, ugly domes on top of everyone's cars. They're way too expensive, and even if the price can come down, they're way too prone to adverse weather. Wherein, what, do you tell people that the car can't drive? Or that they have to?

      I think the future on this front is time-of-flight cameras as a supplemental datastream to multi-input driving systems. A company like Tesla could simply swap out their existing cameras for time-of-flight ones and get a LIDAR-like datastream, which they can plug in as a better alternative to their current photogrammetry + radar streams when weather allows for it, and otherwise fall back on photogrammetry + radar. Time-of-flight cameras still yield the same sort of visual information as regular cameras, but can also record the length of time between a IR pulse is issued and when different parts of the CMOS/CCD register the reflection, and thus produce a high resolution "LIDAR" map instantaneously with each pulse throughout the camera's entire field of view. Because they're manufactured with traditional semiconductor methods, in theory they should be little more expensive to produce when in mass production than conventional cameras.

      Either way, though, you still need conventional cameras for object identification, for seeing colours, and a whole host of other things. A z depth map on its own isn't enough.

      --
      The most exciting phrase to hear in science, the one that heralds new discoveries, is not âEureka!â(TM), but
    9. Re:Toyota is... by TheCastro1689 · · Score: 2

      Currently, there are about 218 million drivers and in 2016 there were 6,296,000 police-reported motor vehicle traffic crashes. So that's about a 2.8881% crash rate. Just saying.

    10. Re:Toyota is... by Immerman · · Score: 2

      Why? If it's safer than 90% of drivers, then that 90% would be better off in an autonomous vehicle. The remaining 10% would be better off driving themselves, but even they'd benefit from the rest of the traffic being autonomous - fewer idiots to deal with.

      --
      --- Most topics have many sides worth arguing, allow me to take one opposite you.
  2. I really don't understand the interest here by DeplorableCodeMonkey · · Score: 2

    Slip ups on the road can become fatal in seconds because of the speeds and forces involved. You know people are going to rely on these systems precisely when they should be off the road period or be paying attention. And what happens when a deer decides to bolt out from the woods in front of your vehicle? Are you going to trust that the car can detect a deer?

    This seems like a solution to people who hate the idea of mass transit and transporting goods by trains. Self-driving cars and trucks and hyperloops! FFS, just hire Disney's engineers and building a fucking monorail in most cities and connect them to the suburbs. That would be more than sufficient to raise the quality of life on transit.

    1. Re:I really don't understand the interest here by michiganbob · · Score: 2

      And what happens when a deer decides to bolt out from the woods in front of your vehicle? Are you going to trust that the car can detect a deer?

      Actually, yes. I trust a computer system that can handle thousands of computations every second much more than I trust a startled, panicky human in that scenario. I don't know if you've ever hit a deer with your car, but they don't exactly give you much warning, regardless of how much you're paying attention to the road.

      Source: Lives in Pennsylvania

  3. Caution is important by GWBasic · · Score: 3, Insightful

    I'm a little skeptical of a sudden mass takeover with autonomous driving. As this post implies, the risk is huge. Where are autonomous devices in low-risk situations? Why haven't they taken over? I think we're better off with things like dryers that can sort and fold laundry, or dishwashers that can put the dishes away. The risk of a dropped dish or torn shirt is much more tolerable than a car crash at highway speeds.

    --
    No, I will not work for your startup
    1. Re:Caution is important by hipp5 · · Score: 2

      The risk of a dropped dish or torn shirt is much more tolerable than a car crash at highway speeds.

      But the rewards are also so much lower.

      The potential rewards of autonomous driving are HUGE. I won't pay more than a couple hundred bucks to have my dryer fold laundry. But I would pay a lot to not have have society ever have to face drunk drivers again. To give old people the freedom to get out and about. To have cars that can precision park themselves, thereby taking up way less parking area (no door opening space needed). And so on.

    2. Re:Caution is important by atrex · · Score: 2

      Fabrics are actually very difficult for robotics to deal with because it bunches, snags, etc. It's why the manufacture of clothing is still mostly done in sweatshops instead of being completely automated.

      A dishwasher could be automated, but you'd be talking about having it integrated into a cabinet system and having grippers on slide tracks trying to grab non-metallic, non-magnetic plates and glasses with just the right amount of force not to break them and organize them throughout the cabinet. Any potential convenience would be grossly out weighed by the cost to develop or even manufacture and install the unit.

      There is a ton of automation already out there in "low risk" environments. The average consumer doesn't see it because the cost of it limits it to factories, shipping centers, and other large facilities.

      Arguably, the "high" risk nature of driving is one of the main reasons behind autonomous vehicle development. The developers believe that they can and will ultimately create self driving vehicles that are safer than the average human. And keep in mind that the average of human drivers includes those who are sleepy, sick, intoxicated, staring at their phone, road raging, etc.

      Of course, the other big factor in the research and development of autonomous vehicles is the shipping industry. How much can corporations save by replacing truck drivers with computers? Computers that don't need to sleep or eat and can be on the road traveling from point A to point B 24 hours a day 7 days a week. That's the jewel at the end of their rainbow.

      Unfortunately, there isn't going to be any sudden shift to fully autonomous vehicles, no matter how safe they may end up becoming. The change over is going to be long and arduous, simply because not everyone can just run out and buy a new car. Nor can most existing vehicles be easily retrofitted. I'd wager we're looking at a 30-50 year time span for the development of the vehicles and supporting infrastructure to finish and for them to trickle out through society.

      There's also a large potential for fully automated vehicles to cause disruption in individual car ownership. Cars may not be something everyone owns in 20-30 years. Instead, entire fleets of automated vehicles run by Uber, Lyft, or another startup could have their services provided to individuals via subscription fees and shuttle them here, there, and everywhere all with a few smartphone swipes.

  4. Loong hand-over times by DrTJ · · Score: 4, Insightful

    Toyota is not the only one deliberating skipping L3 and go directly to L4. Volvo intends to to the same, as well as some of the German vendors.

    The reason is that studies show that hand-overs do not only take "a few seconds" according to the article, but that there is a tail of up to 40 seconds before the "driver-to-be" comprehends the situation.

    Since 40 seconds is an eternity in traffic, it poses essentially the same challenges as L4 systems. So why bother with L3?

  5. I'm sick of hearing this BS by Viol8 · · Score: 2

    "Humans are on average really bad drivers"

    No, we're not. We're actually bloody good at it considering we never evolved to drive something weighing 1.5 tons at anything up to 10 times our maximum running speed alongside other vehicles doing similar speeds.

    People cite accident statistics as if they're significant. When you consider the TRILLIONS of miles driven every year by the worlds drivers and the number of potential accidents that DIDN'T happen because drivers reacted properly, the actual number of accidents is a statistical blip.

    If you think computers are so much better at these sorts of tasks than humans then you might like to consider why Airbus computers will hand over control to the pilot if they're unable to handle the situation. And these are computers in a $200,000,000 aircraft, not a $20,000 car!