Latest Tesla Autopilot 2.0 Update Tested – 32-Minute Video

3 months ago by Steven Loveday 32

DeRage has a reputation for providing long, detailed videos for our enjoyment. This one makes you feel like you are actually behind the wheel of a Tesla, utilizing Tesla Autopilot 2.0.

Tesla Autopilot 2.0

Tesla Autopilot 2.0 – Auto Lane Change

The camera is mounted in a driver point-of-view location, and provides a crystal clear view of exactly what is happening on the road, on the instrument cluster, and on the touch screen. DeRage provides a continuous commentary regarding exactly what he is doing, and how the car is responding. Fortunately, unlike some of the other videos that we have seen, the audio is crystal clear.

He assures that the Autopilot (Autosteer) does, in fact, function over 55 mph, and everything seems to work properly as he travels up to 75 mph on Lake Beuna Vista highway in Florida. He also uses the Auto Lane Change features multiple times. DeRage does note that the instrument cluster is not always showing the other vehicles. He says that he assumes the vehicle is using more than one camera, because the car no longer attempts to incorrectly exit or change lanes.

DeRage’s overall impression is that the Tesla Autopilot 2.0 with the newest update is very stable. He explained that now, drivers are just waiting on some of the automated convenience feature, like automatic windshield wipers.

Video Description via DeRage on YouTube:

Just received the latest update which is version 17.11.3 or 8.1 for my Tesla Model S AP2.0 HW2.0

This update bring the list of features much closer to the original AP1 hardware with Summon and higher autosteer speeds on highways.

Tags: , , , , , ,

32 responses to "Latest Tesla Autopilot 2.0 Update Tested – 32-Minute Video"

  1. Terawatt says:

    Having seen the Bolt navigate San Francisco on its own, with pedestrians and intersections and loads of cars all dealt with without incident or need of intervention, and then the recent Fully Charged episode where the LEAF is running around in London all on its own, I have to say it is difficult to understand why so many keep saying that Tesla has such a huge lead in autonomy! While they are certainly at the front of what you can use on the road today, and therefore also racking up the miles (much) more than anyone else, it just doesn’t seem to me like there is any solid reason to believe they actually have better technology..!

    With Intel’s takeover of Mobileye, and Tesla’s switch (loss?) of supplier, it seems to me Tesla could conceivably be at the wrong end of the self-driving spectrum…

    1. DJ says:

      Blasphemer!!!!!!!!

      Tesla was actually at the forefront of what was out there, well once you put restrictions around it like in a full sized car, at full speed, etc.. It’s a good system for what it can do however I just think the limits of the hardware they chose to implement will come and bite them in the ass later on when there are more players on the road. I just don’t see how Tesla’s hardware can compete long term against LiDAR.

      1. floydboy says:

        What is the ACTUAL superiority of LIDAR over vision?

        1. pjwood1 says:

          I think LIDAR, being laser, gets its feedback at the speed of light. Costs $5k-10k. Correct if wrong, but isn’t radar like a sonic frequency (or dopler?), which is to say a lot slower? Cameras and radar are the Tesla recipe. AP2 is a “more=better” approach. When LIDAR achieves economies, I don’t think there’s anybody with know-how who thinks it won’t take over.

          As with current batteries, Musk proved the commercial success of something others won’t touch. He may refine the use of camera inputs, but AP2 owners know how long it took for their cars to catch up with AP1 (Mobileye).

          “He says that he assumes the vehicle is using more than one camera, because the car no longer attempts to incorrectly exit or change lanes.”

          FWIW, AP1 significantly reduced its habbit of exiting, over time. When it can’t see lanes, generally I can’t either. So, I don’t see how one, or four(?) cameras make a difference.

          1. Ambulator says:

            Radar came from (probably) RAdio Detection And Ranging. So, light speed as well. Tesla also uses ultrasonic detectors, which are limited to the speed of sound. At the range Tesla uses them the speed difference is not important.

          2. Pushmi-Pullyu says:

            “…isn’t radar like a sonic frequency (or dopler?)”

            No. You’ve been confused by the terms “Doppler radar” and “the Doppler effect”; the latter usually refers to sound, altho it can also refer to light from distant galaxies.

            RADAR, or RAdio Distance And Ranging, uses radio waves for active scanning, and thus works at the speed of light rather than the much lower speed of sound.

            I’m not an expert in this field, but according to my understanding, perhaps the biggest difference between the way radar and lidar work is that radar uses a spreading beam for scanning, whereas LIDAR (LIght Distance And Ranging) uses a narrow, focused laser beam. In theory at least, lidar should give more accurate scanning since there is almost no spread to the beam, and thus less bounce-back of “ghost” echoes from nearby objects as there is from radar.

            From Wikipedia:

            A Doppler radar is a specialized radar that uses the Doppler effect to produce velocity data about objects at a distance. It does this by bouncing a microwave signal off a desired target and analyzing how the object’s motion has altered the frequency of the returned signal.

        2. Hank S. says:

          The superiority is that you get an actual 3 dimensional space. 2D Vision is subject to many kinds of depth perception errors like the “white trailer against a white sky” issue in the accident that resulted in the driver’s death.

          However, Tesla uses RADAR so a better question would be “what is the superiority of LIDAR to RADAR”. It’s not a subject I’ve looked into, but my understanding is:

          PRO: RADAR can pass through objects while LIDAR would be susceptible to errors from heavy rain, snow, fog or even dirty lenses.

          PRO: RADAR can also tell you the relative velocity of objects

          PRO: Solid state, inexpensive chips

          CON: Generally lower resolution

          CON: Doesn’t work as well for very short distances…thus sonar.

        3. Pushmi-Pullyu says:

          floydboy asked:

          “What is the ACTUAL superiority of LIDAR over vision?”

          1. LIDAR doesn’t care whether it’s day or night, light or dark.

          2. LIDAR scanners, like radar, can provide an accurate and exact range to other objects. Individual cameras, obviously, cannot. Even triangulating using stereo cameras only gives an estimation, at best.

          3. Active scanners such as lidar detectors can provide positive scanning, which is much more reliable than trying to use software to interpret the images seen by cameras. Remember what Tesla said about that fatal accident involving Model S controlled by Autopilot/AutoSteer? They said the Autopilot confused the white side of a semi trailer with the brightly lit sky. That’s the sort of thing that can happen when you rely on software attempting to interpret camera images. Contrariwise, active scanning with LIDAR or RADAR gives a positive feedback to the sensors, which is much more reliable.

          4. Infrared band LIDAR can see right through fog.

          Now, that said, I don’t know that there is any real advantage of LIDAR over RADAR for this application. Tesla seems to be doing quite well at pulling an astonishing amount of data from radar return echoes. I didn’t believe Tesla could really use RADAR to “see” two cars ahead by bounding the radar beam off the road, underneath an intervening vehicle, but they have demonstrated that they can indeed do that. I dunno how reliable that is, but the fact that they can do it at all is something I find astonishing.

        4. needa says:

          Mapping. And using that data with what it sees currently.

    2. Mark.ca says:

      Their technology is used now on the road by owners, the others are just testing it…so let’s compare what actually is there to compare.

    3. floydboy says:

      What the heck is the “wrong end”! From what I’ve seen from the YouTube videos, each iteration from Tesla appears to show substantial improvement in actual customer cars. When I see similar videos of Bolt customers doing the same, I’ll be just as impressed.

    4. RPadTV says:

      It is difficult to understand because you’re not thinking about price and availability. While expensive, you can buy an AP2 Tesla now. Nissan’s first autonomous prototype Leaf was released in 2015 and cost the company $1-million. Who knows when a consumer-ready product will be available. As for that autonomous Bolt you saw, when can you buy it and for how much?

    5. AlphaEdge says:

      But, with the huge number of vehicles sold with the huge amount of data collection, that deficit that Tesla currently has will be massively corrected in short order.

      That Leaf driving itself was amazing, but when sold in cars, when gathering millions of miles of data? By the time they get there, Tesla will be far ahead by that time on the software/AI side.

      Tesla takes a risk, by putting their tech out there early, but it’s the data collection that’s so important, and so it pays to get the hardware out early, and have the software catch up later.

    6. needa says:

      I tend to agree. Once the annual reports came out in Cali on the number of times autonomous vehicles had to be taken over, you can see how far behind Tesla’s system is.

      They only put their cars (5) on the road in Cali for a month while trying to make that video they did, and had hundreds of takeovers on each car.

      I don’t remember the wordage for searching out the data…so cant post links. But there is a post on this site from a couple of weeks ago (google .2 times per thousand miles) where I went into details on it, with links, if anyone wants to see what I am talking about.

      Bottom line, I quickly realized that Tesla is waaaay behind. No matter how many miles they gt each day for simulation.

    7. Pushmi-Pullyu says:

      “I have to say it is difficult to understand why so many keep saying that Tesla has such a huge lead in autonomy! …it just doesn’t seem to me like there is any solid reason to believe they actually have better technology..!”

      Obviously this is a subject on which reasonable people will disagree, because we are not all going to agree on what makes a certain tech “better” or “worse”. That’s a subjective valuation, not an objective one.

      Arguably Waymo’s autonomous car project (recently spun off from Google) has more advanced technology, since they are already using some cars with no steering wheels or other human driver controls. Tesla, so far as I know, ain’t doing that.

      But where it really counts, in my opinion, is where autonomous or semi-autonomous driving systems are saving lives, or not. According to the NHTSA, Tesla cars which have AutoSteer installed have an almost 40% reduction in accident rate. And of course, some of those cars are not going to have the system switched on, which means the increase in safety from actually using AutoSteer is significantly greater than 40%!

      There is no question that Tesla’s AutoPilot/AutoSteer tech is saving lives in the general public, not just making it safer for drivers especially trained to drive for testing programs such as Waymo’s project. And Tesla’s semi-autonomous driving tech is saving lives today, not tomorrow. Can any other company make the same claim? So far as I know, the answer is “No!”

      1. needa says:

        I don’t think that 40% figure can be used for 2.0. It still lacks some key safety features.

        1. Pushmi-Pullyu says:

          Even assuming that’s true today, how long do you think it will be before it’s not?

          If the current version of Tesla Autopilot isn’t already more advanced than the last version using HW 1.0, then it soon will be!

          1. needa says:

            They are obviously having some issues or it would be better already. I would like to think that HW2 will at least match HW1’s capabilities when they show off the final build of the Mod3.

            When I think about how much trouble they had getting a video made for full autonomy, then you throw in how far out they are on HW2, I start to wonder if they are going to be able to make that cross-country trip this year.

    8. Nix says:

      Terewatt, are those production versions of the Leaf and Bolt that are live in customer’s hands? Or are those test cars? Because Tesla has test cars that they drive around in fully autonomous mode too.

      If you are comparing actual live software and hardware in customers hands, to test cars, I think that’s a mistake.

    9. Cavaron says:

      I got the impression that the Bolt and Leaf where driving on a carefully tested track.

      I would say that Tesla has the lead because it already brings short to 10.000 autonomous casr per month on the road. Sure, they will need software updates – but once they are ready, there will be a huge fleet of them already on the streets.

      The biggest benefit is Teslas data mining with these cars. Belive me – they are creating digital road maps in ultra high “resolution” with optical, radar and sonic data. Thats like studying chemistry by memorizing every book that has ever been written about the subject (like brute force deep learning).

      Also I prefere the Tesla sensors, because radar and ultra sonic sound are not limited by things like fog, rain, snowstorms…

      1. Pushmi-Pullyu says:

        Radar is scattered by heavy rain, just as lidar is, altho my understanding is the degree interference isn’t the same for both.

        That’s not to say that radar and lidar sensors are of no value in heavy rain, but the signal-to-noise ratio is compromised. But then, the “Mark I eyeball”, the human vision system, also has its “signal” degraded in heavy rain.

        However, and again if my understanding is correct, radar works just fine in light to moderate rain. I think lidar also works fairly well in light rain, but I dunno about moderate.

        * * * * *

        Now, about those ultrasonic sensors: Those are of use only at very short distances, which means the car will have very little time to react if another car gets close enough for the sensors to detect it. Very little time to react means not enough time to take action to avoid an accident.

        Ultrasonic sensors will be fine in conditions where all vehicles are moving at slow speeds; say, in a parking lot. I suppose they also work well enough in conditions where all vehicles are moving close to the same speed, as is usually — but not always! — the case on the freeway. However, ultrasonic sensors are wholly inadequate in any condition where another vehicle may be approaching at speeds of, say, 30 MPH or more. That can even happen on the freeway in certain conditions: When the semi-autonomous car is accelerating in a merging lane, when it slows down or stops, or when a “speed demon” comes up from behind.

        Ultrasonic sensors are also pretty useless on roads with two-way traffic, because oncoming traffic will be approaching much too fast for those short-range sensors to be of any use.

        1. G. Richard Raab says:

          Ultrasonics can see 8M. That is plenty to help at low – middle speeds. It can also help at high speeds with cars that are beside you.

          The 5 of the cameras see 80 M, while the 3 forwardlooking cameras see various depths of 60, 160 and 235 M.
          the forward looking radar sees 160 M.

          These combined with the NN will allow for changing the code to better suite the edge cases.
          With AP1, it was at the end of its line. Basically, any other improvements was going to costs a great deal more than this, and would be inflexible.

    10. G. Richard Raab says:

      Multiple approaches to doing autopilot.

      The first is using simple heuristics, which is what Tesla did and all others are doing. The problem is that it can cover the corners with any reliability.

      OTOH, Tesla has switched over to having a general purpose set of sensors and camera that acts as a feed to a neural net. So, where else is that used? Why by life, esp. Humans. We have a number of sensors, namely our eyes, but we also use hearing, along with feeling on the roads, and even in our eyes multiple types of vision(day vs night).

      In the case of Tesla, they have added a number of different type of sensors all around so that they can accurately judge where vehicles, tractors, deer, bikes, pedestrians, etc are. Likewise, they have not only the sensors and NN, but also input via the maps. IOW, we all know that Tesla downloads where roads are.

      Combined with the NN (and GAs back at Tesla office), these vehicles are figuring out a great deal much more than simple discrete logic combined with simple heuristics.

      One last thing that is supposed to be coming, is car2car communication. That will make some interesting changes.

  2. georgeS says:

    Hmm. It will only work on the interstate??

    1. Pushmi-Pullyu says:

      No.

      To the best of my recollection, Tesla is claiming Autopilot/AutoSteer will work on any paved road with clear lane markings, altho it works less well on winding roads and roads with sharp turns. That includes roads with two-way traffic.

      The upgrade reported here is Tesla cars equipped with the new Autopilot hardware 2.0 being able to use AutoSteer up to 80 MPH; previously it was limited to a lower speed… perhaps 55 MPH, if I recall correctly? And not much earlier it was limited to just 35 MPH.

      1. georgeS says:

        “To the best of my recollection, Tesla is claiming Autopilot/AutoSteer will work on any paved road with clear lane markings”

        That’s my recollection as well….but the driver in this video seems to say otherwise.

        1. Pushmi-Pullyu says:

          The message he posted to YouTube, accompanying his video, starts with:

          “Just received the latest update which is version 17.11.3 or 8.1 for my Tesla Model S AP2.0 HW2.0…”

          Tesla quite clearly said that Autopilot use at the newly enabled higher speeds — that is, freeway speed — requires calibration which can take, if I recall what Tesla says, up to a week, and must be performed on clearly marked roads.

          If this guy “just got” his upgrade, then it hasn’t yet been properly calibrated for use at freeway speeds.

          I recall many posts to the Tesla Motors Club system from drivers using Autopilot/AutoSteer HW 1.0 before it had been properly calibrated; some were rather hair-raising. Bottom line: If calibration of Autopilot HW 2.0 works like calibration of HW 1.0, then all Teslae drivers had better remain ready to grab the wheel at any moment, until the system has properly calibrated itself. And unfortunately, I don’t know that the system ever alerts the driver to say “Okay, now I’m calibrated”… so there may be drivers out there thinking their system has calibrated itself when it hasn’t.

          * * * * *

          I admit that none of this makes any sense to me. Why does every car have to be calibrated independently? Why can’t Tesla just perform the calibration using their test cars, then include that calibration in the next software update? No doubt there is a good reason for this, but speaking as a computer programmer I find it baffling.

    2. needa says:

      I doubt they will release it for other types until they get AEB up to snuff.

  3. David Drake says:

    Why is that Semi doing around 70 MPH? If it is loaded isn’t that extremely dangerous and reckless?

  4. Martin Winlow says:

    I am amazed that IEVs is promoting videos showing Tesla drivers completely flaunting Tesla’s ‘always hold the steering wheel’ requirement when using Autopilot. I appreciate if you are demonstrating a semi-autonomous car that taking one’s hands off the wheel might be necessary for a short time but this is the most irresponsible sort of journalism.

    And this guy is complaining about other drivers around him holding cell phones being scary…

    Unbelievable.

    1. Pushmi-Pullyu says:

      If Tesla ever said that drivers should “always hold the wheel” when using AutoPilot, they certainly don’t any more. The last official Tesla video I saw which demonstrates semi-autonomous driving showed the “driver” sitting with his hands on his knees. Unlike Tesla’s previous videos, he didn’t even have his hands palms up ready to grab the wheel in an instant.

      1. Pushmi-Pullyu says:

        Correction: My comment refers to AutoSteer, not Autopilot.

Leave a Reply