Elon Musk called lidar a ‘crutch,’ but now Tesla is reportedly testing Luminar’s laser sensors

[TruckElectric]

Elon Musk called lidar a ‘crutch,’ but now Tesla is reportedly testing Luminar’s laser sensors

Bloomberg confirmed that the two companies have a partnership

By Andrew J. Hawkins@andyjayhawk May 24, 2021, 2:19pm EDT

A Tesla Model Y was photographed in Florida sporting rooftop lidar sensors made by buzzy sensor manufacturer Luminar. The sighting caused a bit of a stir among Tesla watchers, given Tesla CEO Elon Musk’s well-established disdain for the laser sensors commonly used by autonomous vehicle companies to create 3D maps of their environment.

Even more notably, Tesla has reportedly entered a partnership with Luminar to use lidar for “testing and developing,” according to Bloomberg. What exactly this partnership entails we don’t know for sure — neither company is commenting. But it could point to some shortcomings in the technology Tesla is using to power its “Full Self-Driving” driver assist feature.

The vehicle was spotted last week in Palm Beach, Florida, by Grayson Brulte, a consultant for the AV industry who lives in the area. After Brulte tweeted the photos of the Model Y with a rooftop rack of lidar sensors on May 20th, Luminar’s stock surged to its highest level yet. (The Florida-based company recently went public after being acquired by a “blank check” special purpose acquisition company, or SPAC.)



Luminar sold the lidar to Tesla as part of an agreement between the two companies, according to Bloomberg. Also, the Model Y was sporting a manufacturer’s license plate that is registered to Tesla in California. The same plate has been spotted on other Tesla vehicles, including a prototype of the Cybertruck.

It’s unlikely that Tesla will reverse its position on lidar based on a single vehicle. As Guidehouse’s Sam Abuelsamid told Bloomberg, it’s more likely that Tesla is using Luminar’s lidar to validate its Full Self-Driving feature. But it’s still a noteworthy development given Musk’s vocal animosity toward the sensor. In a 2018 earnings call, Musk said, “In my view, it’s a crutch that will drive companies to a local maximum that they will find very hard to get out of.” He added, “Perhaps I am wrong, and I will look like a fool. But I am quite certain that I am not.”

“PERHAPS I AM WRONG, AND I WILL LOOK LIKE A FOOL. BUT I AM QUITE CERTAIN THAT I AM NOT.”
Then a year later, he called lidar “a fool’s errand” during a presentation on Tesla’s efforts to build fully autonomous vehicles. “[A]nyone relying on LIDAR is doomed,” he added. “Doomed. Expensive sensors that are unnecessary. It’s like having a whole bunch of expensive appendices... you’ll see.”

Musk said Tesla is trying to tackle a much bigger problem: passive optical recognition. This is why Tesla is banking on cameras as the key piece of hardware for autonomous vehicles. With their ever-increasing pixel resolution and the low price point, camera sensors are seen as indispensable for advanced driver assistance systems (like Tesla’s Autopilot) and fully autonomous systems. For Tesla, cameras are everything.

Musk also walked back some of those comments in recent months. In a chat on Clubhouse, Musk admitted to “talking smack” about lidar but noted that SpaceX — his other company — has developed its own versions of the laser sensors to assist the Dragon capsule. In a recent earnings call, Musk spoke about moving away from using radar, stating, “We believe that a vision-only system is ultimately all that is needed for full autonomy.”

Luminar, which is based in Florida, went public last year via a reverse merger with a SPAC. That merger valued the company at approximately $2.9 billion in “implied pro forma enterprise value,” with an equity value of $3.4 billion at closing. Luminar is also working with Pony.ai, Airbus, Volvo, Audi, and Toyota Research Institute.



SOURCE: THE VERGE


Tesla Tests Luminar Sensor Musk Scorned as ‘Fool’s Errand’

Ed Ludlow, Gabrielle Coppola and Dana Hull


(Bloomberg) -- Tesla Inc., whose boss Elon Musk says he can develop a fully self-driving car without using laser-sensor technology, has a contract to use such sensors from Luminar Technologies Inc. for testing and development, according to people familiar with the matter.

A photo of a Tesla Model Y with a roof rack of lidar sensors posted to Twitter on Thursday stoked speculation that the companies were working together, sending Orlando, Florida-based Luminar’s shares up more than 10% that day. While similar images of Teslas with lidar sensors have popped up before on social media, it was unclear until now what the relationship is between Tesla and Luminar.

Luminar sold Tesla the sensors on that Model Y as part of an agreement between the companies, some of the people said, requesting anonymity because the information is private. In addition, Bloomberg News confirmed with the California Department of Motor Vehicles that the manufacturer’s license plate on the photographed vehicle belongs to Tesla. The same manufacturer’s plate has been spotted on other Tesla vehicles, including a prototype of the forthcoming Cybertruck.


Luminar rose in early trading Monday. A Luminar representative declined to comment. Multiple Tesla executives contacted by Bloomberg didn’t respond to a request for comment. Tesla has disbanded its media relations department.

The Model Y with the sensors was spotted in Palm Beach, Florida, around 8 a.m. Thursday by Grayson Brulte, a consultant to the autonomous-vehicle industry who lives in the area. The photos he posted to Twitter fanned market speculation that Chief Executive Officer Musk might be hedging his claim that he can develop a camera-only self-driving system, without the radar and laser sensors embraced by competitors.

Luminar’s shares pared a gain of as much as 9.8% to trade up 4.7% to $22.42 as of 9:44 a.m. in New York. The stock had surged on May 20 to the highest level in more than three weeks and its ticker symbol, LAZR, was shared widely on Twitter in speculation that it had supplied the sensor suite.

It’s not clear what Tesla’s intentions are with Luminar’s lidar, the people familiar with the matter told Bloomberg. Tesla engineers could be benchmarking their camera-centric self-driving system against lidar; they also could be testing lidar for future use.

Any suggestion that Musk would change course to adopt lidar is unrealistic, said Sam Abuelsamid, an analyst with Guidehouse Insights.

“The more likely scenario is they are using the Luminar lidars to validate their camera-based FSD system,” Abuelsamid said in an interview, using an acronym for full self-driving. “If they made that change, it would effectively deprecate their entire fleet of vehicles. They are not going to retrofit one million vehicles.”

Key Technology

Lidar, which stands for light detection and ranging, is a key technology for the world’s leading automakers, tech companies and startups as they race to make fully autonomous cars a reality. Lidar uses lasers to build a three-dimensional image of the surrounding landscape, which is critical to how the vehicle perceives the environment, predicts the behavior of pedestrians and other vehicles, and plans how to safely navigate roads.

Musk has been a lone, and loud, voice saying that lidar isn’t necessary.

“Lidar is a fool’s errand,” he said during Tesla’s “Autonomy Day” for investors in April 2019. “Anyone relying on lidar is doomed.”

Luminar, which went public in a $3.4 billion reverse merger last year, has production agreements to supply lidar sensors to Volvo Car AB and the flagship R brand of SAIC Motor Corp., China’s biggest automaker.

Read more: With Robotaxis Still a Distant Dream, Lidar Makes Itself Useful

More recently, Musk has softened some of his public comments on lidar technology. In an appearance on the audio chatroom Clubhouse earlier this year, he admitted to “talking smack” about lidar and said that SpaceX -- the rocket company Musk also leads -- developed its own internal lidar to aid the SpaceX Dragon capsule. “Obviously if I hated lidar I wouldn’t have done that,” Musk said. “However, for driving on real world roads you have to solve vision, understanding objects.”

On an earnings call last month, Musk suggested Tesla is also moving away from radar, which, along with lidar and cameras, is one of the three technologies the industry has viewed as critical sensor hardware needed to reach a future of self-driving vehicles.

Analyst Adam Jonas of Morgan Stanley called that a bold and risky move. Jeff Osborne of Cowen & Co. went a bit further: “We continue to question Tesla’s approach to vehicle automation,” he wrote in a note to clients. “In our view this is unlikely to succeed.”


SOURCE: YAHOO FINANCE

Sponsored

 

[Crissa]

You can't make definitive statements like Musk makes without testing.

If you don't test, you're just talking out of your hat. If you do test, you know.

-Crissa

 

[TruckElectric]

 

[TruckElectric]

Teslas With LIDARs And No Radars — It’s Not What You Think

Point cloud from a Luminar LIDAR demo

BRAD TEMPLETON

There’s been buzz about a few recent Tesla TSLA +4.4% news items this week:

1. A Tesla factory car was spotted with a Luminar long-range LIDAR on it
2. Folks noticed Tesla’s ordering page now brags about the car’s long-range vision capabilities and not about radar, which Tesla has said they want to stop using.

It’s well known that Elon Musk has declared LIDAR a “crutch” he will never use — he says it distracts from the real mission of doing it all with computer vision. He’s probably, but not certainly wrong, and is placing a big bet on getting the breakthrough needed to make it all work with vision.

The sighting of a Tesla with a LIDAR doesn’t mean they’ve reversed course, though. It doesn’t even mean they are investigating it, though that would be a reasonable idea.

Tesla wants to solve the problem with machine learning. Without LIDAR’s fully reliable way to detect the distance to targets in your field of view, they have to train neural nets to figure out distance from images. You typically train machine learning systems by showing them lots of example of correctly interpreting what you want the system to learn. For example, you take an image and draw a box around a car to tell it what a car looks like, and you may also tag the box with a distance, to let it know how far away the car is.

If you do that enough times, the system can learn how to figure out distance from the image. This is what human beings do, though we are pretty clever at it, and we don’t yet know how to make computers do it quite so well that you would bet your life on it, and bet your life is what you must do for a full self driving system. A real, actual full self driving system, that is.

The above method, with the tagged images, is called “supervised learning” and it’s a lot of work to tag all those images. People thus hunt for ways to train computers without all that labor. One way to do that is to have another way to know the “ground truth” — the reality — of an image. So if you go out with a LIDAR, it can make images where it knows the distance to every section of the image, and produces training data.

This technique is well known, but Tesla has also been trying something that goes even further. The holy grail is “unsupervised” learning, where the system is able to figure out patterns without humans telling it what they are. That’s powerful, but difficult. Tesla is researching an approach called “self-supervised.” In that approach, the system tries to guess right answers, and it uses clues to figure out if that guess is correct or not.

When it comes to figuring out how far away a car is, you work from a video, not a single frame. In a series of frames, you might guess each time how far away your target is. But the “self-supervision” comes from the fact that you know something about physics. Real objects don’t just randomly teleport. If your guess in frame 1 was 100 feet and in frame 2 it was 150 feet, then at least one of those guesses was wrong. If your guess in frame 1 is 100 feet, in frame 2 is 101 feet and in frame 3 it’s 102 feet, this is far more likely to be right. So you strengthen your network there and make it like that guessing approach over others. And you do this a billion times, and you can get somewhere. It doesn’t matter you had to do it a billions times because the machine did all the work.

Tesla has been touting that approach for some time. The appearance of the LIDAR car suggests they are probably not giving up on the ground truth approach that comes from the LIDAR. Or they may just be using it to test their system to see if it can predict the same as the LIDAR. If they can always match the LIDAR at figuring the distance, with no errors, maybe they can reach that reliability you would bet your life on, and do it without the cost of the LIDAR. Or so Tesla hopes.

Radar

An employee holds a radar sensor main circuit board on the production line at the Continental AG ... [+] © 2018 BLOOMBERG FINANCE LP

Teslas have used radar for some time. Radar, like LIDAR, gives you a very reliable distance measurement, and also a speed measurement, which is great. Problem is, you get much less information about where the radar target is. The radar may roughly tell you it’s in your lane, but that’s about it. And it may be bouncing off something in the environment and not in your lane at all, though it looks like it is. There’s a lot of noise in the radar world, especially from signals bouncing off things that are not moving, like the road, and the planet, and the signs and bridges. There’s so much of this that you often just have to ignore any radar signals from things that are not moving towards you or away from you.

That’s a shame, because while it’s great to spot all those moving objects like cars (even through the fog or rain) you would like to see stalled cars, and trucks crossing the road. Famously, Tesla cars in Autopilot crashed directly into such things when their drivers stopped paying attention. In several instances, leading to the death of those drivers. You definitely can’t bet your life on it.

So far, Tesla’s efforts at computer vision in the Autopilot product have also failed to do the job and spot things like cars at an angle partway into their lane, or even big trucks. The systems just didn’t get trained on a lot of those and weren’t able to identify what they were and how far away they were in a reliable way. You can’t brake for ghosts all the time, so you have to be really reliable at this, and Tesla’s not there yet. Radar and LIDAR are reliable at this, except for those problems listed above for radar. This has made people wonder why Tesla is suddenly being so quick to try to work without it.

Speculation online has wondered if the semiconductor shortage might play a role. It is known that many thousands of Tesla cars are sitting in factories, almost finished but missing some key parts. Maybe one key part. Tesla doesn’t say what it is. So people wonder if it’s a radar part, and wonder if it may be some time before these radars can be had. That might trigger an emergency attempt to rewrite Autopilot to work without radar.

While Tesla’s logic about LIDAR applies to radar to some degree — Tesla says that you need superb computer vision to drive, and once you have that, you don’t need LIDAR any more. But radar has some tricks up its sleeve that you would not want to give up:

1. You see radar bounces from cars ahead of you that you can’t see because they are blocked by the car right in front of you. You can tell when those cars hit the brakes before you see them, before the car in front of you hits them. That’s valuable.
2. Radar sees perfectly well in fog and rain, which LIDAR and cameras don’t.

LIDAR works with the same lightwaves that cameras do. If you can use a camera and get distance, you don’t need LIDAR. Radar is different, and it tells you things no camera can ever tell you. It’s also not that expensive. It doesn’t make sense to give it up, not in the race to be first at making self-driving work.

A new generation of radar is arriving, sometimes called 4D or “imaging” radar. This radar has much more resolution. It doesn’t just tell you what lane you are looking at. Some hope for radars that see almost as good as LIDAR with 0.1 degrees of resolution, though nobody is there yet. Such radars would be very useful to have. Tesla has been rumored to talk to ARBE, an imaging radar company from Israel. It makes sense, and it doesn’t make sense for Tesla to have no love for radar even if it declares LIDAR a crutch. The truth is right now, computer vision only has one leg, and needs that crutch.


SOURCE: FORBES

 

[SSonnentag]

Likely nothing more than a research and/or calibration device.

Nothing to see here in my opinion.

Sponsored