With New Autopilot Update, Tesla Model S, X Much Improved On Highway – Video

3 months ago by Steven Loveday 28

Videos prove that the latest Tesla Autopilot 2.0 update makes for a much improved freeway driving experience, however, navigating local poorly-marked roads is a different story.

As you can clearly see from the video above, Tesla’s Autosteer is still not up to par on local roads. The driver notes frequent times that the Autopilot system wants him to take control. It isn’t seeing lines or curbs, and at one point it nearly drives directly into a center divider. This video is telling because the driver has driven the route before and he directs our attention to certain areas, explaining what happened in the past, compared to what we are seeing happen with the updated Autopilot.

Håkan Forss shares via YouTube:

Testing out Autosteer on my usual test run on a local road with the new firmware version 8.1 17.17.4.
This version is performing very poorly on this local road. The tracking of the lane markers and the curb is much worse than previous versions. It has a hard time making the turns and turns of Autosteer multiple time where it has performed much better in the past.

Not in this video but, this version seems to perform much better than previous on highways and multi-lane roads with good lane markers.

Opinions are mixed, however. The above video shows a Tesla faring quite well on France’s curvy highway roads. Perhaps the fact that it’s a highway, and that the markings are more clear, makes all the difference. The road surface looks new and is dark in color, and the white lines are thick and vibrant. He’s even able to change lanes using Auto Lane Change while in the midst of a curve. Though the video is entitled “French Curvy Road,” and there are surely some wide bends in the highway, it isn’t all that curvy.

The video above shows another Tesla owner on a curvy two-lane highway. The roadway is more similar to the first video, however, again the markings are more pronounced. He isn’t dealing with intersections or high traffic areas. Essentially, it’s a single lane with a clear yellow dividing line, and nothing but farmland off the shoulders. Again, not nearly quite curvy or demanding as the first video.

The takeaway from all of this is that the Tesla Autopilot update seems to be much improved when it comes to highway driving, and on well-marked roads, without lane splits or major intersections. But, as soon as you add such “challenging” situations, it’s clear that Tesla still has some work to do.

Håkan Forss, the driver from the first video up top, headed back out the next day and proved our takeaway from this collection of videos. He shows that the update does, in fact, succeed on local roads with good markings. He shares via YouTube:

In this test run Autosteer works really well for most of the time. Very commutable and smooth in most turns. The challenge are tight turns, blind corners and crests.

Source: YouTube via Teslarati

Tags: , , , , , , , ,

28 responses to "With New Autopilot Update, Tesla Model S, X Much Improved On Highway – Video"

  1. Kent Altena says:

    I get wanting the future, but driving with Autopilot is similar to driving with a teenager. You are just waiting for them to fail. It may make you more aware of your surroundings, but driving on pins and needles is not my idea of fun.

    1. georgeS says:

      Kent,

      It has to be an improvement over no autopilot and someone trying to text and drive.

      1. ¯\_(ツ)_/¯ sven says:

        Unless Autopilot gives a false sense of security that it’s now safe/OK to text and drive, increasing the amount of time the driver spends texting while driving, and lowering the driver’s/texter’s vigilance in watching the road while texting.

        1. (⌐■_■) Trollnonymous says:

          Meh, AP.

          Don’t care for it and IMHO is just stuff that will get outdated quick.

          Tesla should offer a version of cars that does not have AP and one that does and see how many sell of each.

          1. jelloslug says:

            Autopilot is optional on every Tesla sold.

              1. Pushmi-Pullyu says:

                I think he meant using it is optional. The hardware isn’t optional, but enabling the software, or not, mostly is optional. Mostly, but not entirely; some Autopilot features can’t be turned off.

            1. Pushmi-Pullyu says:

              “Autopilot is optional on every Tesla sold.”

              Certain Autopilot features, such as AutoSteer, are user-enabled; they can be turned on or off by the driver, using software switches displayed in menus on the car’s giant video screen.

              But not every Autopilot feature can be turned off.

        2. pjwood1 says:

          A point of agreement, here, is why does Tesla illuminate the lane markers on the dash (blue), which means the cameras can see the car and its relationship to the lane, if it can’t then stay within it? I haven’t been able to figure that out, and think there’s some false confidence there.

          I still love the system, and think AP1 was slam, dunk, worth $3k. I’d have paid close to that for adaptive cruise. While I trust the car is more apt to stay in lane, when “blue”, I know it isn’t when it flickers, or goes to grey (cameras don’t see, in heavy rain, snow). Consistency is mostly there, on highways. Stop and go has become more, and more reliable, with better, stronger stops when neeeded.

          1. Pushmi-Pullyu says:

            The blue lines are not where the lane marking painted on the road actually are; they are just where Autopilot “thinks” they are.

            Edge recognition using camera images and optical object recognition software is not perfect, and from what I’ve read, it’s not going to be. The performance of Tesla Autopilot/AutoSteer is unquestionably improving, and improving with impressive speed. But there are hard limits imposed by the fallibility of optical object recognition software. Passive scanning can only provide so much data. You run into what programmers call the “GIGO” problem: Garbage In, Garbage Out. If the program gets bad data, then it doesn’t matter how good or how sophisticated the program is. Bad input produces bad output, in the same way that a logical argument based on false premises usually leads to a false conclusion, regardless of how perfectly logical the argument and its conclusion are. (I’d say inevitably leads to a false conclusion, but sometimes random chance produces the correct conclusion from false reasoning. One entertaining example of that is the “She’s a witch!” scene in “Monty Python and the Holy Grail”.)

            Active scanning, with radar and/or lidar, provides much more reliable data.

    2. Bacardi says:

      The two places you want it the most is stop and go traffic and highways…

    3. Mikael says:

      That is a part of being at the forefront of technology.
      And if you use it at stretches you know it can handle, like parts of your commute, then it is easier to relax a bit.

      If you want a perfect self-driving car then you will just have to wait and be part of when the majority get into this technology.

    4. digicool says:

      You quickly learn when you can trust AP and when not to. Even when I only trust it 80% of the length of the drive, I will not buy another Tesla without it.

      I put the car on AP when its going to be a boring stay put drive. Thats often 2/3rds of long road-trips. Minor corrections to keep the car inside the lane could create a lot of fatigue which AP takes away. Hands on the wheels ready to take over is very different level of involvement compared to making constant minor corrections.

  2. ¯\_(ツ)_/¯ sven says:

    I predict that Elon will eventually eat crow and admit that a camera and radar only approach won’t cut it for Autopilot, and that Lidar will be needed.

    1. pjwood1 says:

      I predict that, just as he did with batteries, Elon would go LIDAR before others would jump on the cost curve. Right now, that’s too high, so he stubbornly believes cameras will evolve.

    2. scottf200 says:

      Lidar is not usable in rain, fog, etc so those systems would need to rely solely on radar just like Tesla. Cameras and radar work more consistently.

    3. Pushmi-Pullyu says:

      I dunno, radar may prove to be better. I did a bit of research on lidar vs. fog the other day, and discovered that even near-infrared scanners, which can see thru fog, need cooling systems. But radar sees thru fog just fine without needing any special equipment.

      Tesla seems to be doing an amazing job at pulling more data out of radar reflections than I thought possible, such as bouncing radar off the road to “see” two cars ahead.

      I think it’s unquestionable that Tesla needs radars facing all points of the compass, not just front-facing radar. In the slowed-down version of Tesla’s “Benny Hill” video (second video at the link below), demonstrating an upgrade to AutoSteer, you can see the very large number of false positives in detecting obstacles to the sides (green outlines). That shows just how fallible depending on cameras is. I personally would not want to trust my life to something with that high a rate of fallibility!

      With active scanning, either radar or lidar, those false positives should almost entirely disappear.

      http://insideevs.com/tesla-releases-self-driving-demonstration-with-recognition-feed-video/

  3. agzand says:

    Sorry but this looks terrible. I think Tesla lost the edge they had with autopilot once they got dumped by Mobileye.

    1. georgeS says:

      agzand,

      In the first video where it performed poorly he said that the earlier version performed better.

      Was the earlier version gen 1 hardware?

      1. agzand says:

        I am not sure, but I guess that is what he means.

        1. protomech says:

          Since he is talking about “new firmware version 8.1 17.17.4”, presumably he means an earlier firmware performed better on local roads.

      2. VazzedUp says:

        Just software, and it may be Tesla migrating from Mobileye to a full Tesla solution currently.

    2. Pushmi-Pullyu says:

      “I think Tesla lost the edge they had with autopilot once they got dumped by Mobileye.”

      Based on all the InsideEVs articles on the subject, it seems rather clear that Tesla has had a struggle to regain the functionality it had in its camera + optical object recognition software, which it had with MobilEye equipment.

      But altho it is taking some time, Tesla continues to make improvements with the new system (Autopilot hardware version 2, aka “HW 2.0”). In some ways HW 2.0 already outperforms HW 1.0; in other areas, they’re still working to reach the former level of functionality. But I’m sure that in time, they’ll fully surpass MobilEye’s system.

  4. (⌐■_■) Trollnonymous says:

    I’m not interested in AP. As a matter of fact, they can omit it all together if there was an option on my future purchase.

    Hands have to be on the wheel, you need to be alert and ready to take action etc…..

    No thanks.

    1. Mark.ca says:

      Completely disagree with you on this…
      I want to be able to activate AP and watch that stupid Harry Potter movie from start to finish without interruptions, ok?

      1. Pushmi-Pullyu says:

        Okay, but doing that safely is going to require reliable, fully developed Level 4 autonomous driving functionality. Tesla is still working on maxing out Level 3, and even Waymo has only partially functional Level 4 tech. (Warning: That is a broad oversimplification. I have no doubt there are certain aspects of autonomous driving where Tesla’s cars perform better than Waymo’s.)

        http://www.techrepublic.com/article/autonomous-driving-levels-0-to-5-understanding-the-differences/

  5. Kalle says:

    Hey! I know that road 🙂

Leave a Reply