Tesla Removes Ultrasonic Sensors In Bold Move That Cripples Features But Promises To Restore Them

Transportation Tesla Removes Ultrasonic Sensors In Bold Move That Cripples Features But Promises To Restore Them Brad Templeton Senior Contributor Opinions expressed by Forbes Contributors are their own. I cover robocar technology & previously worked on Google’s car team. Following New! Follow this author to stay notified about their latest stories.

Got it! Oct 17, 2022, 07:00am EDT | New! Click on the conversation bubble to join the conversation Got it! Share to Facebook Share to Twitter Share to Linkedin A Tesla before removal of the sensors getty Tesla announced that they are removing the ultrasonic sensors from models 3 and Y, and later from S and X. New buyers of these cars will find this means it disables features like the useful auto park and park assist, as well as the less useful summon and smart summon features in the “enhanced Autopilot” and “FSD” packages. While these features will be disabled in these new cars, Tesla promises it will fairly soon re-enable them with a future software release which will provide them using just the cameras.

Park assist (the beeps and screen displays which show you are getting close to walls or other vehicles when parking) and auto-park are now fairly common checklist features for middle and high-end cars. Customers expect them, so removing them is no small step. On the other hand, Tesla is selling all the cars they can make and so probably doesn’t fear it will lose sales over this.

Even so, this opens up many questions: Tesla will save money removing this hardware, but it’s fairly cheap hardware so not a great deal. But why remove it before the replacement software is ready, depriving buyers of a checklist feature? Can Tesla easily implement these features (and more) just from cameras, and if so, why isn’t this working already? How does this compare to the way Tesla removed radar from new cars, reducing functionality, then slowly (but not completely) brought it back? How does this illustrate Tesla’s car-as-computer approach in contrast to other automakers? What are ultrasonic sensors? The ultrasonic sensor in the bumper of a Tesla San Francisco Chronicle via Getty Images Ultrasonic sensors are cheap, small “sonar” that sends out an inaudible sound wave, and wait for it to reflect back off things at the speed of sound. They can tell you how far away any significant object is out to a few meters.

Tesla has 12 of them, mostly on the bumpers. These cost only a few dollars though mounting them and wiring them adds to that. Munro Live estimates the total cost at about $114.

Almost all automakers uses these to provide park assist warnings and more, as did Tesla. They give a very simple picture of the world — is there anything out there and how far away is it? Some fancier ones generate a low resolution image. They are also used for blindspot warnings and they can detect the guardrail next to the road.

MORE FOR YOU Juan Soto Contract Rejection Could Make Orioles A Better Buy Than Nationals Paris Car Show Sheds Most International Input, But China Turns Up In Strength Nascar Great Jimmie Johnson’s ‘Reinventing The Wheel’ A Home Movie Of His IndyCar Season The picture is pretty poor, so many people find park assist to be annoying, providing beep warnings too early or on things you are not on trajectory to hit. Many prefer the use of a backup camera, at least for going in reverse. But the ultrasonics make the job easier.

They are mainly mounted in bumpers, which does increase the cost of accidents and insurance, however. With Tesla’s sales of 40,000 cars/month, $114 amounts to under $5M per month, which is not chicken feed, but also not a lot of Tesla thinks it will restore parking assist in just a few months. Once they make the software do it from cameras the decision is a no-brainer.

When Tesla pulled radar it was commonly speculated to possibly be due to the global parts shortages — some alleged Teslas were sitting ready to ship except for the radar, and pulling radar fixed that. When they pulled radar, the new cars had a much lower speed limit on Autopilot, longer following distances and many users reported a serious increase in the amount of “phantom braking” — when the car brakes for a ghost, a ghost that radar might have confirmed wasn’t there. Over time, Tesla still hasn’t restored the full 90mph Autopilot speed it had with radar, and phantom braking is still an issue.

The currently 85mph speed is pretty decent, though the 80mph they had for a while was not sufficient on faster highways where that speed, while illegal, is still pretty common. Even so, radars are more expensive than ultrasonics, and they were hard to find. That doesn’t seem to apply here.

Tesla’s autopark originally used only the ultrasonics and could not see things like lines in a parking lot — it needed to have cars on either side to park between. Later it evolved to use both the cameras and the ultrasonics. Can they do it? Tesla is following an all-computer-vision approach, and in particular has recently rolled out a testing version of their “virtual LIDAR” which attempts to build a 3D representation of the world just from camera data, without necessarily being able to identify things to figure out how far away they are.

(The human mind uses its knowledge of how big cars and people are to guess the distance to them. ) They are now calling this the “occupancy network” system, an effort to build a map of what parts of the world before the car are “occupied” and which are empty, thus allowing the car to decide where it can go. Tesla’s old diagram of sensors on the car.

The 12 white dots show the location of the now removed . . .

[+] ultrasonic sensors Tesla If you build such an accurate 3-D map of the world, it’s then pretty easy to provide parking assist and auto-park within it. In fact, you should be able to do a better job. Typical ultrasonic based park assists don’t really understand the shape of things and give alerts about things the car is not about to hit.

A better system can predict where the car is going to go and know when a warning is needed, or park with more confidence. The problem is that there are areas the cameras can’t see, most notably the space right in front of the car, which is invisible to both the driver’s eyes and the camera. To provide parking assist here, the car must map the space when it can see it, and work from that memory.

That can work fine as long as nothing in the space is likely to move. Fortunately the most likely things to move — other vehicles and pedestrians — are all tall and are visible to the camera even when their bottom half can’t be seen. A small child might be hidden, but a system should identify that and simply not operate if there is a small child in front of the car, obviously.

Cameras also have a limit to their depth of field and can’t easily deal with things very close to them, but once again it is sufficient to map the objects when they are not that close and remember their shape. If they move, that can be detected, even if blurry. Providing parking assist is one thing.

Self-driving may be more difficult due to the blind spots of the cameras. A parked car that wants to leave on its own may not see an obstacle that has appeared in front of it that is not visible to the forward cameras. Many robocar designs have LIDARs on stalks to make sure they can see all around the vehicle, which a Tesla can’t easily do.

For driver assist, the human can be counted on to fill this gap. The ultrasound sensors also provided an assured perception of things like guardrails when on the highway, which gives more confidence to Autopilot and self-driving functions. Even if the perception system were to make a mistake about the lanes or guardrails, the ultrasonics have assured the vehicle won’t veer into the guardrail under any circumstances.

This redundancy will be gone. Software Car Most interesting is how this illustrates how different a car company Tesla is from the old-school OEMs, even now that they woke up about the Tesla threat many years ago. Tesla is probably going to get away with removing the parking features from the car, though it may take a short-term hit on sales of enhanced Autopilot (for whom the parking features were the best part) and FSD.

It’s hard to imagine old OEMs saying, “Oh, sorry, your car won’t have that feature, but give it some time and it will reappear with a software feature. ” Customers would not trust them and probably wouldn’t have it. For many, the fact that Teslas regularly get new features after you buy them has been a nice selling point for the cars, though the rate of new features has slowed a fair bit in the last few years with all the focus on FSD eating resources.

Customers will be willing to accept Tesla’s promise that they will get this feature before too long. Even in spite of Elon Musk’s current poor track record on promising dates to deliver functions. In the computer and phone world, it’s pretty normal to buy hardware and expect it to get important new functionality later with a software update.

That’s not been true with cars. Follow me on Twitter or LinkedIn . Check out my website .

Brad Templeton Editorial Standards Print Reprints & Permissions.