#1 cleantech news, reviews, & analysis site in the world. Subscribe today. The future is now.

Autonomous Vehicles

Published on July 11th, 2016 | by Michael Barnard


Tesla Autopilot Had 1st Fatality Because It’s Better

July 11th, 2016 by  

Screen Shot 2016-07-06 at 7.30.20 AMTesla’s Autopilot is in the news these days with the tragic death of a Tesla owner while Autopilot was in use. However, there are at least five competitors with technology that they assert works similarly to Autopilot, many of whom have been on the roads longer than Tesla’s offering.

Why hasn’t the inevitable happened with them? One frequently promoted hypothesis is that it’s because their technology is better. A more credible one is that it’s because their technology is quite a bit worse, so is much less used by drivers.

Screen Shot 2016-07-08 at 9.08.39 AMLet’s start with the question of which other companies have similar capabilities on the road today. Car and Driver magazine tested offerings from four companies back to back: Tesla, Mercedes, BMW, and Infiniti. Motor Trend tested Cadillac and Hyundai capabilities against the Tesla and Mercedes offerings as well.

There are numerous semi-autonomous capabilities on most modern cars that we take for granted. Some are ones that the National Highway Traffic Safety Administration (NHTSA) and National Transportation Safety Board (NTSB) in the US and similar organizations in other countries have strongly advocated for over the years due to their impressive safety improvements. The table below provides a summary of the various capabilities and which manufacturers provide them.

Screen Shot 2016-07-10 at 7.59.27 AM

The first observation that can be made is that Tesla offers more features than other companies, but all six companies offer numerous driver assist features. Cars are assisting drivers much more than most people realize these days.

Tesla is somewhat differentiated by offering this technology on all of its cars, but as it only has two at present, this isn’t a major point. More important is that the other manufacturers offer the more advanced features only on their most expensive cars, and it’s not available on their mass-market offerings. In theory, the Cadillac has lane following with its current level of technology, but in practice, it worked so poorly that it is unusable in its current version per Motor Trend.

What did Car and Driver say about the various systems?

Well, on the key metric related to Autopilot, it’s much better than the other systems.

Screen Shot 2016-07-06 at 7.32.41 AM

autonomous-driving-route-inline-photo-665750-s-originalThe key row is the number of lane control interruptions on a 50-mile route. They drove each car across a 50-mile (80-km) route involving highway and rural roads as well as the main street of a town. This is a decent representation of average roads in the USA, disrepair and all, including rainy night driving.

The Tesla handled far more of the scenarios than any other vehicle. While others required driver input every mile or more on average, Tesla was closer to double that, handling a greater number of conditions with greater ease.

The sedan that begs to differ is this test’s clear winner. With utmost confidence and only two cautions from legal counsel—“Always keep your hands on the wheel. Be prepared to take over at any time”—the Tesla Model S locks onto the path ahead with a cruise missile’s determination and your hands resting on your lap.


The Tesla’s Autosteer performance can be distinguished from our other contenders by two words: no wobbling. This car identifies the exact center of your lane of travel and holds that course with minimal deviation. This system rises well above parlor-trick status to beg your use in daily driving.

That’s an important point. Tesla’s offering is so superior that it actually is trustworthy while the others per back-to-back testing aren’t. What is trustworthy is used, which is possibly part of the reason why Tesla publishes how many miles have been driven under its more advanced features and the other car companies don’t. (Another possible reason is that Tesla knows how many miles it has been used for due to its approach to the automation, and the other car companies don’t.)

What about Motor Trend?

Autonomous-Testing-Route-Map-freeway-vs-canyon-roadMotor Trend tested a lot of different conditions, including cars cutting into the lane, cars braking in front of the car, acceleration when traffic speeds up, and very importantly, how many times people had to intervene over two different types of driving roads: “a 35-mile stretch of an immaculately maintained 65-mph toll road and a windy 17-mile, 55-mph two-laner.”

What did they find on the key question of how usable the advanced features were?

motortrend car interventions

There’s little comparison. The Cadillac was so poor that it couldn’t be considered a real offering, and the Tesla is literally an order of magnitude better than the other cars under real-world conditions in one test. The differentiation from competitive offerings is even more pronounced in Motor Trend‘s tests than in Car & Driver‘s.

Tesla’s current production capabilities in two sets of automotive journalist testing outshone their competitors. While some competitors have shown impressive capabilities in artificial demonstrations or non-production research work, a production Tesla today per Motor Trend “is nearly a Level 3 car.”

Is this due to the magazines being biased sources, as someone insisted to me recently? Given that Tesla spends $6 per car sold on advertising compared to competitors’ thousands of dollars per car on their premium models, it’s hard to imagine that any bias exists in Tesla’s favour with the mainstream automotive press.

This level of sophistication and competence leads to drivers using it. A large part of the reason Tesla is in the spotlight for the fatality is because its driver assist features are so much better than its competitors. Simply put, fatalities are a statistical inevitability of driving, but the other manufacturers’ technology hasn’t been used enough that the sad inevitability has occurred yet.

Perversely, the better autonomous driving technology gets, the more deaths will occur while it is being used. Tesla is merely at the leading edge of this reality.

Tags: , , , , , , , , , , , , , , ,

About the Author

works with startups, existing businesses and investors to identify opportunities for significant bottom line growth in the transforming low-carbon economy. He regularly publishes analyses of low-carbon technology and policy in sites including Newsweek, Slate, Forbes, Huffington Post, Quartz, CleanTechnica and RenewEconomy, with some of his work included in textbooks. Third-party articles on his analyses and interviews have been published in dozens of news sites globally and have reached #1 on Reddit Science. Much of his work originates on Quora.com, where Mike has been a Top Writer annually since 2012. He's available for consultation, speaking engagements and Board positions.

  • Steven F

    In regards to the accident that killed the Tesla driver. many people assume the driver had enough time to respond break and avoid the accident. There is another possibility. The driver may have not had enough time to react. The maximum legal length a truck with trailer in the US ia about 67 feet. The Tesla hit the trailer just behind the truck drive wheels and and went under the trailer. About 30 feet from the front bumper.

    No if we assume the truck stopped and then started his turn he probably would have been moving about 10 miles an hour at the time of impact. At 10 miles an hour the truck would cover 14 feet per second. This would give the driver about 2 seconds to respond . Being optimistic maybe 3 seconds. The driver would have to respond immediately to avoid hitting the trailer. If his reaction was a little slow he would hit the truck.

    However it we assume the truck didn’t stop, but instead slowed to 30 miles and the driver quickly checked for traffic (but didn’t see the Tesla) and then made the turn (a rolling stop) The Tesla driver would have had about 1 second to react. That is not enough time to stop.

    Based on my guesses the driver would have had at most 3 seconds to respond and could have had less than 1 second. If he was speeding as many people have asumed. the tesla would have been traveling between 100 to 130 feet per second. Even if the driver responded immediately he probably didn’t have enough time to stop before hitting the trailer.

  • Bob_Wallace

    Can someone who took geometry within the last half century (I’m 55 years out and don’t want to spend hours refreshing) help me out with vertical resolution for lidar?

    Here’s what one system claims it can do…

    “The ULTRA Puck doubles the range and resolution (via number of laser channels) of its predecessor to 200 meters and 32 channels, providing enhanced resolution to identify objects easily. The 32 channels in the ULTRA Puck are deployed over a vertical field of view of 28° and are configured in a unique pattern to provide improved resolution in the horizon to be even more useful for automotive applications. ”


    32 channels over 28° would mean the ability to detect less than 1° at 200 meters (~650′)?

    The bottom of a trailer is about 4′, the standard minimum for an overhead sign is 17′. Would that difference be detectable at 350′ (within the stopping distance of a T85 going 90 miles per hour)?

    Can you calculate the point between 200 meters and 0 meters at which the lidar system would provide the car’s computer enough information to determine that the height angle was not increasing as it would with an overhead sign?

  • john

    Absolutely correct. 50 million miles. One fatality. These features depend on only two things. Image processing algorithms and forward facing radar. Tesla just set their height of observation for their radar to low. That will be fixed in the next automatic software update. Image processing is solely dependent on processor capabilities. Since Tesla platforms have by far and away the fastest processors it is the most advanced.

  • steve garside

    Tesla may simply be offering features that the other manufacturers have decided are not within current capabilities.
    Safety-critical aerospace, medicine, oil&gas, all have standards and rules that must be followed. In those fields, hardware may appear to be overburdened to the commercial eye. ‘Hey, you don’t need all that stuff, we can do that with Arduino and a USB camera’
    It’s easy for a Maverick to ignore the rules and methods, and the experience that produced them; that is, until something goes badly wrong.

    • To be clear, the automotive industry is a safety critical industry as well with very extensive and long-running embedded system with safety engineering.

      All of the capabilities listed in the first table are embedded software systems with a strong safety focus. Almost all of those vendors have released variants of the Autopilot technology under question.

      Mobileye visual technologies are used by multiple manufacturers and radar is used by several as well.

      Asserting that Tesla is a maverick in this case is not well supported by the evidence. And the testing shows that Tesla’s solution is much more robust than the also production-released solutions from other manufacturers. It seems as if the mavericks are the older manufacturers.

      • Bob_Wallace

        I want to see testing of the other collision avoidance systems to see if they can detect a white trailer against a bright sky.

        I would think one of the car magazine/sites would be all over this.

      • steve garside

        Cars are subject to safety regulations for already mature systems. Autonomy is not in that group, but largely self-regulated. Other manufactures, whether using Mobileye or not, are more conservative in promotion, and offered features.

        The journalists’ tests are vague and subjective, but say the Tesla is better
        at lane following.
        Better lane tracking can come at the expense of speed of response, or limited forward detection. As with RADAR used to track incoming missiles, extended range and sensitivity can increase the occurrence of ‘false positives’. In Naval applications, waves produce ‘noise’, making threat detection more difficult.

        Similarly, road signs, trees and lighting, all produce noise. Limited range and sensitivity reduce the detection of the possible. The car manufacturer may make the choice; ignore what the system can’t handle, or pass control to the driver. If the former fails, blame the driver.
        Tesla may provide the driver with a more comforting experience, but a less capable one, where collision avoidance is the priority. It is Tesla’s failure to detect a collision that is the matter at hand, not how many times the steering wheel is touched.

    • Greg Hudson

      One also might consider that the Israli (Mobileye ?) camera system does not apparently produce a ‘normal’ color image like a gopro (or other digi cam). The weird images I’ve seen are barely recognizable as a picture !

  • harisA

    Although information presented is useful, but the title has a contradiction ( I would use the term oxymoronish if I were trolling:-))

    • No contradiction. Tesla’s Autopilot is good enough that a lot of people use it a lot of the time. As a result, the statistical inevitability of a fatality fell to them. The other manufacturers’ offerings are expensive toys for bragging rights, not actually useful systems that people would use. As a result, they don’t get used and in not being used don’t see accidents and the unfortunate fatality. It seems contradictory, but it’s not.

      • Greg Hudson

        I beg to differ… Mercedes trucks have been testing for years, and are now driving in Europe on selected routes.

  • beernotwar

    We shouldn’t miss that the cause of this accident wasn’t entirely due to the failure of Tesla’s sensors to detect the side of the truck. Why didn’t the driver detect the side of the truck? Because he was watching a movie. In other words he trusted the autopilot to such a degree he was comfortable letting it do all the work.

    This is both good and bad and brings up a huge issue for this tech going forward. To the good, Tesla’s tech is good enough to convince people to trust it. To the bad, people misuse technology all the time. We will need to learn how to use driverless tech and how not to. Integrating this tech into the dynamic and infinitely variable real world is not always going to go smoothly when people start making bad assumptions about its capabilities. Which they will. People are going to have sex in the back of a driverless car with no one at the wheel. It is going to happen. There will be a name for the club you join when you do this. And when someone crashes in flagrante delicto it will be another scandal.

    People are, as usual, likely to be the weakest link in our bright, driverless future.

    • Bob_Wallace

      We don’t actually know what the driver was doing. We’ve heard one report that he was watching a movie. And, IIRC, the police report stated that the DVD player and laptop were turned off.

      There was also a report that he was watching a movie on the car’s screen, which seems to be impossible.

      But you’re right. Humans are the weak link. There’s no excuse (that I can think of) for a human driver to not see the tractor cross the lane and the rear wheels of the trailer. Even if the trailers was painted the exact same shade of blue that was in the sky behind.

      • Also the police report put the truck at fault for failure to yield

        • Bob_Wallace

          Most likely the final judgement will share the blame.

          Even if you have the right of way you do not have the right to plow on through. You have a responsibility to avoid (or at least attempt) to avoid the crash.

          • Agreed.

            I’d prefer to be at fault and alive rather than not at fault and dead. Just because the truck driver was at fault doesn’t mean its OK for others to let a collision to occur.

          • Tesla reports the driver and the Autopilot were blind to the truck so if they paint him some blame it will probably be too fast for conditions. The truck will get failure to yield or failure to clear the intersection before proceeding and who knows what else.

      • Greg Hudson

        Hey Bob… Maybe it (the truck) had stealth technology aka like what Top Gear in the UK created on day with a bevvy of screens on one side, and a camera on the other? It was not bad for amateurs.

  • Bob_Wallace

    Forward collision detection automatic braking is clearly not adequate.

    I wonder if the other cars from the table are able to detect a trailer across the road.

    • Impossible to say what edge conditions they fail at simply because they are so little used.

      • Bob_Wallace

        This would be a very easy test to perform. If the other manufacturers have not already done this test after learning about Tesla’s failure then they are negligent.

        • Hard to say. They’d have to replicate the specific low contrast situation with no side rails. It’s an unusual situation and no regulatory testing framework requires it. I’m sure that there’s some bench testing going on, but putting an expensive car on a track on a very sunny day and running it at a white tractor trailer would be challenging to justify for many of them. Not saying that they shouldn’t do it but I suspect that they are assessing the technical capabilities of their solution compared to these conditions and making a determination without having to run their cars at solid objects.

          As Mobileye has stated that this is a condition their technology currently doesn’t resolve well, any manufacturer using Mobileye is definitely sitting up and taking notice.

          • Bob_Wallace

            No, just assume the camera system would fail. Check to see if the radar/lidar system would save the car and occupants.

          • GCO

            You mention the lack of side rails as a possible reason why Mr Brown Model S rode under a trailer.
            Are you suggesting that it would have been okay to drive at full speed between the wheels of a moving truck?

          • No but on Brown’s Model S the radar is in the lower grille vs on the current models it’s somewhere around the bumper or headlight level. So it literally sees a gap with some fuzziness on the upper bound that matches the signature of an overhead sign. Trailers are significantly high off the ground.

          • Bob_Wallace

            Trailers are about 4′ off the ground. Overhead highway signs are supposed to be at a minimum of 17′ off the ground.

          • when you’re 6-8″ off the ground 5-6′ vs 17′ look about the same if you only see reflections and use you seeing partner (the camera) to verify what you sense

          • Bob_Wallace

            From 400′ away can your eyes not tell if something is 4′ off the ground vs. 17′ off the ground?

            Mine can.

          • Mine can too, but the camera was blinded by glare, the radar module can only tell if it receives a signal back and the one that came back from that far away looked like a road sign when cresting a hill or similar

          • Bob_Wallace

            At 50′ feet away the signal would have been dead ahead of the car and not up in the air. There’s no indication that the car activated the brake system.

          • correct the software installed on his car wouldn’t fire the brakes if only one sensor saw an obstruction, 50 feet is beyond the range of the ultrasonics and the camera was blind, so only one sensor

          • Of course not.

            The top of the Tesla and the driver’s head and likely shoulders were sheared off by the bottom of the trailer. Side under rails are mandated in the EU. The system most likely would have detected them and braked and any impact would have resulted in the defence in depth of the crumple zones and airbags coming into play.

    • Greg Hudson

      Trailer or other object, no matter what it is, relying on forward looking optics should not be foremost (if it is). Ideally RADAR should be the primary method of detection, and it would have no trouble detecting a bloody great slab of metal in front of it. IMO there’s more to this story than meets the ‘MobilEye’…

      • Bob_Wallace

        I’m looking forward to an official reporting of what happened.

        If the system was getting continuous information from the radar that there was something out front it seems like the system would be programmed to alert the driver and slow down/stop. I can see tossing out the odd bit of info now and then because you will get some false readings. But this must have been a continuous data stream over time.

        • Greg Hudson

          Maybe this accident might spur on the adoption of a road based version of ADSB as used in modern aircraft. Yes, it would take quite a while for it to be implemented, but it could be done. Hobby Disclaimer: I receive aircraft ADSB data via a specially tuned antenna (1050Mhz) and a small data decoder. Process it on my PC (for my local airspace) and upload the data into a central database in Sweden. The data is then available to look at via the FlightRadar24 web site, giving near real time tracking of aircraft fitted with ADSB transponders.

  • Andre Needham

    One correction in the 4th paragraph: NTSB stands for “National Transportation Safety Board” not “…Traffic…”.

  • crevasse

    And Mercedes announces they’re going to put a prototype competitor out there in the future. Prototype+future means next decade. I guess that gives them time to get their autopilot up to par, but in 4 years the bar will have improved significantly. Sucks being asleep behind the wheel (in a Mercedes).

  • Frank

    One other point not really brought out, is even assuming you are using the feature on one of the other cars, if it needs your help every mile, you aren’t going to take your eyes off the road and watch a movie. You aren’t going to trust it that much.

    • I agree up to a point. I just don’t think that the other solutions get any use to speak of as they are so inferior. If they don’t inspire sufficient trust, they won’t get used at all.

      • Bob_Wallace

        Crash avoidance systems should always be on. All cars that have them.
        The Volvo (?) lane-keeping system operates only at low, stop and go, speeds.

  • Ivor O’Connor

    Thank you for the article. I did not know how the Tesla compared to other vehicles.

  • Brunel

    Cars in Europe are required to have AEB and can detect pedestrians while this Tesla could not detect the side of a truck!

    Maybe AEB should be tested with bright lights pointed at the cameras in Subaru cars.

    • jramskov

      Nice trolling.
      AEB is not a requirement in the EU. To get top safety result in Euro NCAP testing, it is a requirement, but that doesn’t make it illegal to sell cars in the EU without AEB. Furthermore, many of these systems only work at low speed, not at the speed at which this accident happened.

      I do agree that it would be great if it was made a requirement because it would prevent a lot of accidents.

      • Brunel

        How am I trolling!

        80km/h is low speed?


        • jramskov

          Please re-read my post.

          1) You stated it was required in the EU/Europe, it’s not.
          2) You imply that AEB’s in european cars are vastly better than what Tesla includes in their cars without any proof at all.

          I call that trolling.

          Ps. I didn’t state systems that worked at higher speeds wasn’t
          possible/available, I stated that many systems only work at low speeds. 80km/h good, but far from enough. Travel speed is often considerably faster. The speed limit on many high ways here in Denmark is 130km/h, which means it should cover at least up to that speed and preferably higher.

          • Brunel

            The video said AEB is required.

          • jramskov

            No it doesn’t. It talks about HGV’s, not cars and as it says in the end, it’s much more limited than what that Volvo truck is capable of.

      • Bob_Wallace

        That was not trolling. It was a very reasonable observation.

        • jramskov

          Stating that AEB’s are a requirement in the EU when it isn’t?

          • Bob_Wallace

            If you are correct about AEB not being a requirement then the comment would have been bad information, not trolling.

            Internet trolling is something along this line – “make a deliberately offensive or provocative online posting with the aim of upsetting someone or eliciting an angry response from them”.

            What would be good is for you or Brunel to bring some proof re: required/not required.

          • jramskov

            At the same time as stating AEB required in all cars in the EU he’s also implying that it’s vastly better than Teslas.

            Proving it is easy, check more or less any car makers website (for a country in the EU) and you’ll find cars sold without AEB. It’s particularly bad here in Denmark because we have the worlds highest taxes on cars.

          • Bob_Wallace

            Not trolling.

            I’ve got no time to get into the EU/AEB stuff.

          • Harry Johnson

            But you’re the happiest people in the world.

      • Greg Hudson

        AEB may not be a requirement in the EU, but it is about to become law here in Australia (if it hasn’t already). Car manufacturers are not happy about it either…

        • jramskov

          I wish it was law here in the EU.

          • Greg Hudson

            I think it was last year that our Govt mandated that Electronic Stability Control for all new cars. Many of the new cars already here that were not sold had to be returned to their country of origin. There were lots of very unhappy vehicle executives. The same will happen with AEB I imagine. See:

          • jramskov

            Then I would argue the law was badly made. It should be fairly easy to make a law that doesn’t include cars already in the country.

  • JamesWimberley

    Good stuff. The best summary statistic would probably be serious accidents per million.miles driven under software control. Deaths are too rare to be statistically informative.
    It would be nice to include Google’s autonomous cars, even though they are not on sale.

    • Brunel

      Google cars have lidar on the roof.

      Tesla cars have cameras instead of lidar.

      Lidar is superior from a safety point of view.

      • Pilar8542

        I basically make close to 6.000-8.000 dollars monthly from freelancing at home. Everyone willing to work easy computer-based work for 2-5 h every day from your home and earn good income in the same time… This is perfect for you… SELF97.COM

      • MaartenV

        But still too expensive.

        • Modok EvilMastermind

          Interesting though to see some stealth testing of it by Tesla though. Perhaps not too expensive? Or they are just waiting things out until it get cheaper?


          • My assumption is that it is a calibration rig for their three other sensor sets, not something that they are intending to use. Tesla doesn’t make slow cars, it doesn’t make cars with terrible drag coefficients and doesn’t make ugly cars. LIDAR loses on two of those three counts.

          • Bob_Wallace

            The lidar unit on this car is “behind the front panel” which I assume is the faux grill.



          • Interesting tech. That might be worth using. It doesn’t give specific capabilities not met by the combination of radar (differently implemented per another discussion point above), Mobileye visual recognition and sonar, but it is a net new implementation that might be a better compromise set as part of a sensor suite.

            I was wondering about the range question, as the link references 150 meters. 65 mph is 29 meters per second, so 150 meters goes by fast. However, Mobileye’s tech and research make 100 meters the necessary range for most capabilities.

            What’s also interesting to me about Mobileye’s approach upon reading this paper is that it turns out it uses a single monocular camera to achieve its results, not stereoscopic cameras as I had assumed. Other links indicate it’s actually quite low resolution as well.



            Minor upgrades to image processing and tech are required to achieve the necessary visual processing to avoid the low contrast situation being as much of an issue.

            It’s going to be interesting to see where Tesla goes with this. This LIDAR tech makes it a option, and Tesla is very pragmatic. They might go there if the company can scale the tech.

            I don’t see it being necessary, but that doesn’t mean it wouldn’t be a better choice.

          • Bob_Wallace

            I don’t know whether lidar would be better than radar. I have the impression that lidar can build higher detailed images. Radar probably has more range. For a very safe car it might take one of each. One simple radar unit looking forward and slowing the car if there’s a possible problem ahead at 2x the stopping distance. Let the lidar data determine whether the car can or cannot resume speed.

            Cameras are blind in the dark although some sensors are amazingly able to pull and image from light conditions that are difficult for humans.

            Here’s a lidar with 200 meters range. The Tesla should be able to stop in 650′ at greater than 100 MPH speeds.


            We don’t need a lidar puck. We don’t need detailed detection for more than about 120 degrees in front of the car. Side/blind spot and rear problem detection can be done with cameras. Turn on backup lights for the camera when backing up.

          • Karl the brewer

            “I was wondering about the range question, as the link references 150 meters. 65 mph is 29 meters per second, so 150 meters goes by fast. However, Mobileye’s tech and research make 100 meters the necessary range for most capabilities.”

            Whats the reaction time of the software compared to a human?

            Average for human about 0.275 seconds – http://www.humanbenchmark.com/tests/reactiontime/statistics

            So a 0.1s improvement gives 2.9m shorter stopping distance. If software is substantially quicker perhaps the vehicle doesn’t need to see any further.

          • Bob_Wallace

            About 450′. Within the Tesla S85’s stopping distance at 90 MPH.

          • Bob_Wallace

            At 80 mph a car travels 117 feet during the human perception/reaction period.

            “Studies have shown that it takes the average driver from one-half to three-quarters of a second to perceive a need to hit the brakes, and another three-quarters of a second to move your foot from the gas to the brake pedal.”


          • Karl the brewer

            Lets be generous and give humans 0.5 second perception plus another 0.5 second to move, so one second. If the software is 0.5 second faster then that’s 14.5m difference. Difference between life and death?

          • Bob_Wallace

            Autopilot/self-driving systems should never allow the car to drive faster than its ability to safely stop. Including enough space for any following cars to stop and avoid a rear end collision.

            I’d even support a system in cars that reported to police when a car is traveling more than “20 miles”* above the posted speed limit. Then let the police sort out whether it was acceptable speeding (evading someone shooting from another car, rushing to get a seriously injured person to the emergency room).

            Exceed “reasonable speeding” and the lights should automatically start flashing. Horn beeping.

            Want to race? Take it to the track.

            * Maybe 10 miles in a 25 MPH zone, 20 miles in a 65 MPH zone. Some reasonable limit.

          • GCO

            Interesting (and telling) that you’d put aesthetics ahead of safety…

          • It’s interesting that you would respond with this trolling and nasty point well down in a thread of comments which included this explanatory statement by me:
            “The last point to make is that the technology that people will use is the technology that will save lives. Tesla’s choices are unobtrusive and do not mar the aesthetics of the vehicle. LIDAR by definition has to be top-mounted and fairly high above the roof line in order to get a reasonable view of the near sides and front of the car. It’s aesthetically displeasing and while there is zero testing with consumers to find out if they will accept it, I can pretty much guarantee that extremely few people would buy a car with a huge sensor bundle sticking out of the roof. Saying LIDAR is safer is like saying that one life jacket is safer than another life jacket when both are stashed under the thwarts of the boat.”

            That you chose to comment in a snarky way on point that was fully elaborated elsewhere reflects on you, not on me.

          • GCO

            The comment you quote is what pushed me to reply. You seem to agree that cameras can fail in challenging contrast conditions (e.g. white trailer against sky), yet continue to argue against LIDAR because of your preconception that “by definition” (whose??) it can only go on the roof and therefore look bad.
            This is not the case, as @Bob_Wallace:disqus has already pointed out, with references, in this very discussion.

            Had Google and others decided to go with cameras as primary long-distance sensor, and Tesla with LIDAR, would you be making the same argument?

          • You aren’t following the bouncing timeline, troll. If you were, you’d note the discussion with Bob includes my comments on that. Instead, you just seem to think it’s worth commenting nastily on my personal ethics.

            Do you have anything of any merit to add, or is it going to continue to be attempted and kind of pitiful character assassination?

          • Bob_Wallace

            Ease up on the name-calling, Mike.

            This is something that may or may not bear on the discussion –

            Let me point out that there is a problem following the bouncing timeline. I think this is a change that Disqus made a few weeks back, but now if you happen to enter the conversation via an email notification (hit Reply) then you see only what was posted up to the point of that specific comment.

            I’ve had major problems with this. I’ll reply to something posted a “an hour” before and will miss a lot of content that has been posted after.

        • Bob_Wallace

          Possibly not. There are lidar units which could cost under $200 if built at scale. And are about the size of a tennis ball.


      • No, actually it isn’t.

        At one level it’s a superior form of sensor, but safety is a multi-factorial, defence-in-depth issue, not a single sensor issue.

        Tesla’s started with very robust cars, best in class really. They gave them great structural rigidity, better crumple zones front, rear and sides, excellent air bags. They gave them excellent acceleration, deceleration and turning. That’s a primary safety measure.

        For sensors, they have three sets: radar, stereoscopic cameras and sonar. Once again, multiple tools create an overlapping set of defences, as opposed to a single sensor type.

        For automation, they started with avoidance front collision, side collision and rear collision and added additional capabilities for safer driving with the advanced autopilot capabilities.

        This is a fundamentally superior and best practices approach compared to Google’s, which has a single sensor and depends solely on software for survivability.

        The last point to make is that the technology that people will use is the technology that will save lives. Tesla’s choices are unobtrusive and do not mar the aesthetics of the vehicle. LIDAR by definition has to be top-mounted and fairly high above the roof line in order to get a reasonable view of the near sides and front of the car. It’s aesthetically displeasing and while there is zero testing with consumers to find out if they will accept it, I can pretty much guarantee that extremely few people would buy a car with a huge sensor bundle sticking out of the roof. Saying LIDAR is safer is like saying that one life jacket is safer than another life jacket when both are stashed under the thwarts of the boat.

        Saying LIDAR is safer is saying that a single thing makes all the difference for safety when in fact it’s many overlapping things which make something safe.

        • Brunel

          Sonar is for underwater.

          Are you saying Google cars have no cameras and no radar.

        • Bob_Wallace

          If the Tesla has radar then why didn’t it report an object at windshield height 500′ feet away and trigger automatic braking? The Tesla should have been able to stop itself within 300′ to 400′ if traveling 90 MPH.

          IIRC the autopilot will not operate at speeds greater than 99 MPH.

          See my previous comment about the size of lidar that is now being developed. A unit like the one I linked could be installed in front of the rear view mirror where it would do nothing to change looks or aerodynamics.

          • This is an interesting question. The forward radar sensor is at fender level and apparently does not sense anything above the level of the hood. This is a choice that was made given defence in depth with the Mobileye vision technology, a compromise I presume, but I am reverse engineer decision making that seems a little odd to me. It’s not a limitation of radar in general, it’s a limitation of this specific implementation.

          • Bob_Wallace

            I agree on the limitation of implementation. I want to hear why radar did not see the trailer. Radar is not blinded by bright light.

            My working hypothesis is that the team designing the system did not consider having something over the road which was higher than the hood but lower than the roof. And that the fix is to redesign the radar mount.

          • dRanger

            It’s not totally clear yet, but the radar may have seen the truck. Tesla said that both the radar and the camera must agree there is an obstruction in the road before the system applies emergency braking. This avoids false emergency stops which would be a serious hazard. Tesla also said the camera didn’t see the truck so emergency braking would not have been applied. We don’t know for sure yet.

          • Bob_Wallace

            The radar system would have been sending out several probes per second. If the radar system says that there is something solid that you’re about to crash into the car should stop. At the very minimum it should start slowing before it reaches the “panic stop” range.

            The reason to have more than one systems is that so there’s a better chance of picking up a problem. If you’re going to ignore the system that reports a problem and rely on the system that says there is no problem then you’re doing it backwards.

          • dRanger

            The radar can only return a “yes/no”‘signal and it can’t see road markings. What if it sees an object directly ahead, but that object is a car parked along side the road on a curve? The camera must confirm that the object is not in your lane. Radar votes yes – camera votes no and so no emergency braking. As I said before, I find the nuances of the system fascinating and really hesitate to second-guess them without knowing all the details.

          • Bob_Wallace

            This was not a “parked along side the road”. This was a trailer directly over the lane.

            If the car can’t sort out the issue then it needs to slow. Perhaps even pull to the shoulder and stop. The default should not be “Charge forward and determine if there’s a problem by crashing/not crashing”.

            As far as an upcoming curve, your car knows that. Or it should. That’s GPS stuff.

            In this case, Tesla beta, the car should have 1) slowed and 2) alerted the driver. If the driver did not respond then the car should assume the driver is either dead or climbed out a window and safely parked itself.

          • dRanger

            “That’s GPS stuff.” The GPS signal is not inherently accurate enough to use for impact avoidance, nor is it always available, and that car ahead could actually be in your lane. The system can fail by not preventing an accident and it can fail by actively causing an accident. False emergency braking clearly is in the latter category. Imagine the headline “Tesla Autopilot False Sudden Braking Causes Three Fatalities on Interstate” and you get the idea.

          • Bob_Wallace

            I did not say that GPS was that accurate. But GPS is accurate to know if the road is straight or curved ahead.

            I also have stated that the car should not put itself in the position of having to panic stop. If there is less than certainty, slow down. Park if it’s really confusing.

          • dRanger

            I’m really not trying to nitpick here, but no combination of GPS and radar can determine if the car ahead is in your lane or not and this situation is going to come up a lot more often than a white truck pulling directly in front of you against a bright sky. If you disagree, then I guess we’re done.

          • Bob_Wallace

            OK, then move to a lidar system that has the ability to make finer calculations.

          • dRanger

            Yep – as soon as any of the car manufacturers can make that work, I’m all for it

          • Steven F

            Radar cannot create an image of what is ahead. All it can do it tell you something is XXX meters ahead. It also cannot tell you if the object is alongside the road in the middle of the the road or even if the object is above the road. All it knows is that the radio signal it is sending out is bouncing back.

            Now high end military radar are better but they are simply not affordable for a car.

          • Same as lidar at the time this system was designed.

    • Bob_Wallace

      Yes, we need to be looking at all accidents, not just fatalities.

    • On the stat’s perspective, I agree. It’s not an available or reported statistic yet for Autopilot or it’s competitive equivalents. Certainly one of the thoughts I had was that the overall survivability of the Tesla — defence in depth I mentioned elsewhere in comments — would potentially have eliminated fatalities that would occur under Autopilot where some other cars might have seen fatalities.

      Regarding Google, I’ve compared Google’s approach in other places and to be clear it’s not a production system and may never become one. What is interesting in my opinion is that these two sets of tests were of production available solutions.

      • Bob_Wallace

        Tesla is looking at the data. Here’s what Musk had to say –

        “The probability of having an accident is 50% lower if you have Autopilot on. Even with our first version. So we can see basically what’s the average number of kilometers to an accident – accident defined by airbag deployment. Even with this early version, it’s almost twice as good as a person.”


        50% better is not good enough for me. 90% of all accidents (IIRC) are caused by driver impairment (drinking, texting, falling asleep, etc.). The self-driving system is going to do none of that and neither are many drivers.

        The self-driving system has to be better than non-impaired drivers. It has to have less than 10% as many accidents, IMHO.

        • GCO

          Good points. Furthermore, Autopilot can only be enabled on roads which are already inherently safer per km/mile, divided highways, and in sufficiently favorable conditions.
          Tesla never made public an apple-to-apple comparison, because it would likely show that AP as implemented today doesn’t increase but negatively impact safety.

          • dRanger

            “Autopilot can only be enabled…” Not true. Autopilot can be enabled anywhere the system identifies lane markers. It limits speed to 5 mph over the speed limit on undivided roads and Tesla does not recommend its use in town, but the use is not actually restricted.

        • dRanger

          I am curious about your logic here, Bob. If Autopilot had been statistically proven to cut your risk of an accident by 50% which includes the risk of any shortcomings and imperfections it may have, why wouldn’t you turn it on?

          • Bob_Wallace

            Someone who never drives impaired has a fatal accident rate that is 10x below the national average of one per 90 million miles IIRC. If I remember the correct number then someone who never drives impaired has one fatal accident per 900 million miles.

            Tesla now has one fatal accident per 130 million miles.*

            * Until we get to at least a billion miles (according to Musk) we won’t be able to make a meaningful ‘fatalities per miles’ statement for the autopilot. And it might take a lot more than one billion miles depending on how the data plays out.

            If the one per 130 million was a reliable number then we’d have to say that the autopilot is about 7x worse than an unimpaired driver. And it’s worse than that because most of the autopilot data comes from well market, divided highways. That is (IIRC) not where people tend to kill themselves.

          • dRanger

            OK, I understand your position now. You are saying we don’t have enough data to statistically demonstrate that Autopilot is safer than nothing, and on that point we agree. I think even Elon agrees with that.

        • Thanks for reminding me of that. I’m interested in the numbers behind it of course to ensure that it’s actually a valid statistic not just an interesting set of data points.

Back to Top ↑