BMW tests next-gen LiDAR to beat Tesla to Level 3 self-driving cars

L4sBot@lemmy.worldmod to Technology@lemmy.world – 369 points – BMW tests next-gen LiDAR to beat Tesla to Level 3 self-driving cars
zdnet.com

BMW tests next-gen LiDAR to beat Tesla to Level 3 self-driving cars::Tesla's autonomous vehicle tech has been perennially stuck at Level 2 self-driving, as BMW and other rivals try to leapfrog to Level 3.

81

Tesla's decision to only use cameras and no lidar will bite them in the ass.

*Musk’s. He regularly overrules the Tesla engineers.

https://medium.com/predict/musk-overruled-tesla-engineers-and-now-they-are-in-serious-trouble-2e37269e387a

This. That cocksucker has such ~a tiny dick~ fragile ego he makes huge decisions without any expertise simply because he says so. Thats how he built the whole “genius” thing around him. Reality of it is that he is an annoying dumbass who thinks he knows it all and anyone in the same room with more than one brain cell is immediately annoyed with him. But he has a lot of money so i guess LeTtEr X cOoL

I think you need to use two tildes on each side if you want it show up as a strike through like this

 ~~like this~~

Really? On wefwef my comment shows up with a strikethrough

~test with one tilde~

test with two tildes

I’ll use two from now I guess.

I don't know wefwef at all, but on the old Reddit apps, as well as Liftoff and Sync, it seems like two is necessary. I'll check your test on desktop Firefox when I get on my computer next, just out of curiosity

No big deal either way! I seem too remember something similar happening with RIF and spoiler tags, too (I think that one was sensitive to where you put after the opening spoiler tag indicator, but it if you were in "new" Reddit, it didn't matter about the spaces)

Markdown is not portable and its annoying. Why cant we have one standard. Come on ISO

Why cant we have one standard.

*Links to xkcd that everyone always links to

There are different versions of Markdown and they all have slightly different standards although where they're broadly the same they are variable in the particular.

Reddit and I think lemme, although I'm not quite sure, use a version called Cramdown, the problem is if the app rendering the code doesn't think they are using Cramdown and uses something else you get inconsistencies in rendering.

Wikipedia also uses their own media version to be awkward.

Came here to say this. Couldn’t be more on point. Using both cameras and LiDAR in tandem will be necessary for true self driving vehicles.

I've been working with pretty much top tier vision tech, and it's still so far off from being viable enough. It's insane how well it works, but to use it in a dynamic environment driving around on roads... Pure madness to believe it will be enough with only cameras in like the next 10-15 years at least (imo).

I bought a Tesla despite my hatred for Musk because it was the only reasonable alternative when it came to electric cars where I live at the time, and got one of the old ones with lidar and cameras. I have this super weak shimmer of hope that they will go back to using both again, since the warning systems and their "autopilot" feel way shittier than it used to since they've allegedly patched away the lidar.

fortunately LIDAR unit costs are going down, so multiple units, fusing their data with regular camera arrays should resolve a very good view, and be good at error-correcting for each other's shortcomings.

Cameras are better for surveillance of people and can better be sold. Lidar data not.

Currently they seem to be leading the race though even though the competition is using radar and lidar

Edit: Am I wrong?

If buy leading the race you mean the only company to have an actual product available for purchase then yeah.

But the reason they were able to get to market so quickly is because they don't actually have any concerns about it being functional or safe. That's a real boon to them because it helps them move quickly ahead of the competition that do care about those things.

Of course one good argue that an unsafe self-driving system is in fact not a self-driving system and therefore they are not the first to market.

The average consumer would define self driving as "if my car crashes, my car should be sued". Is that how it works with a tesla crash, who pays for that?

It's on the driver's responsibility

Then what's the point in it?

What's the point in a self-driving system that has been babysat in order to ensure it doesn't murder you, random pedestrians and other road users. If I want a car that is unsafe if I take my hands off the wheel I can get a regular car, it already does that.

Tesla themselves call it FSD, Full Self Driving. That is at best false advertising and at worst reckless endangerment. It isn't fully capable, and it requires the driver's attention so it isn't self-driving. Literally every part of its name is wrong.

No, I mean leading the race as in having the most capable sefl driving system in existence which I believe is the case.

I don't know what you're basing the claim on that it's not functional and safe.

I am basing my claim on it not being functional and safe.

I'm basing my claim on the fact that it drives into trucks. Since I don't want to be driven into a truck by my car, I would consider that to be a failure state.

Do some research.

I don't think anyone has ever claimed it's flawless. After all it's still in beta version. If you hit a truck it's because the driver wasn't paying attention.

I still don't know what you're basing these claims on except your own opinion apparently. "It's not safe" compared to what? As far as I know Tesla FSD has had less accidents per mile than an average driver.

In the 2nd quarter, we recorded one crash for every 4.41 million miles driven in which drivers were using Autopilot technology (Autosteer and active safety features). For drivers who were not using Autopilot technology (no Autosteer and active safety features), we recorded one crash for every 1.2 million miles driven. By comparison, NHTSA’s most recent data shows that in the United States there is an automobile crash every 484,000 miles.

Source

Perhaps you should do some more research?

Humans drive using “cameras” (eyes) and no LiDAR, that’s the assumption Tesla is making — that a supercomputer can drive 10x better than humans using the same type of sensor. Nobody really knows yet if that’s true but I get the logic.

LiDAR also is UV/visible spectrum and is blocked by dust/fog/snow/rain so it doesn’t help much in many driving situations..

You're making an argument against LiDAR with it using UV/visible spectrum, guess what uses visible spectrum to see stuff? Cameras. And they also have an unfortunate downside of not having good dynamic range, so in very bright/low light situations they probably don't work that well either. Teslas aren't even using infrared cameras to see in the dark to my knowledge.

Unsure why you are downvoted, because that is sound logic. I recall hearing on a podcast of I think a former Tesla engineer that having too many sensors potentially makes things less effective since you have to deal with different types of input, and have to crunch more data, etc. etc. Efficient development also means knowing when to cull unnecessary time sinks.

I hate Elon as much as the next guy, but... Well, humans are obviously not perfect drivers, but Tesla clearly believes that in time, with cameras all around the car (already an improvement over human drivers), a good enough AI solution would be able to match or surpass humans.

I still rather have good ol radar as a fallback if cameras and their AI model don't work for some reason. They are still work in progress, and rely on trained models to recognize objects, while if a radar sees something, it is because there is something actually there and not a guess. I don't buy the story that too much sensors is bad. Planes rely on multiple different sensors plus backups for redundancy to fly safely, self-driving cars with vastly superior tech should be able to do the same.

They actually have used radar along with the cameras for a while. It wasn't until the last few years that they decided to ditch them. I think I read that they realized it was a mistake and are going to add them back on future models.

Anything short of fiercely anti-Tesla gets immediately downvoted here. Just how it goes I guess. I’m not the biggest Tesla fan but hope they succeed on this front, we desperately need driving assistance technology to make the roads safer.

1 more...
1 more...
1 more...
4 more...

I hate that the article opens with

Just a decade ago, the concept of self-driving cars might have seemed like something out of a science fiction movie

Ten years ago there was already a ton of competition in self driving car research. They were first legalized on the roads 10 years ago. Tesla autopilot (including it even though it was a scam) was sold 9 years ago. Google spun off its self driving car division as waymo in 2016.

This feels like one of those "bruh Zelda ocarina of time came out 29 years ago, we old" memes

Hell, Mercedes and Bosch were testing it all the way back in 1993:

https://www.youtube.com/watch?v=JTnBiTIvGqY

Old Mercedes was insane. They really were the best or nothing

And what happened though, was it unfilled hype or that death caused by that Uber autopilot? Back when waymo was grabbing headlines you would've thought we'd be a lot closer to driver less by now.

This will probably be under monthly subscription

"The route you selected contains a highway. Please purchase the Highway Driving Pack in addition to your City Driving Pack to reach your destination"

Fekkin hell that sounds like a boring dystopia

Almost certainly.

But self-driving also depends on up-to-date mapping data and continually improved algorithms for the autonomous systems to work properly. An ongoing cost to the customer makes the most sense for a service that has operating costs to the service provider.

I mean, does it? Presumably the idea (that Tesla had anyway) is to try and mimic what humans do, and we don't need mapping data to drive "safely" (for a given value of safe). Of course, humans also get lost, but again, the GPS updates is basically free at this point for the mapping help humans need. (Garmin stopped charging yearly long ago, Open Maps and Google Maps and Wayze all are "free").

In the interest of competition, I am very happy that Tesla chose a different path. Self driving is not guaranteed to ever work so we need to try things until at least one works

2 more...

Only if they're giving the hardware away with the car. Tesla is charging ~$15k upfront for FSD. It would be absurd to tack a monthly fee on top of that.

Why do you think you should get stuff for free?

Hardware has a cost. Running a service has a cost. Providing updates has a cost. If you don’t think you should be responsible for the costs of the things you use, you’re going to be disappointed pretty often. Venture-capital funded startups can only give away free shit for so long before they have to start giving returns

No kidding it has a cost, as we're discussing the cost of it here if you're not aware. There's quite a range between 'free' and $15k + $299 a month. You know what else has a cost? The car itself when you purchase it.

2 more...

The goal is than no one owns anything anymore. Every company is after the sweet sweet recurring revenue.

Sample pricing for BMW self-driving add-on feature:

98% accuracy in obstacle avoidance - $299/mo.
85% accuracy in obstacle avoidance - $199/mo.
75% accuracy in obstacle avoidance (lowest legal limit!) - $99/mo.

disclaimer: BMW cannot guarantee 100% accuracy in accuracy rates

Another possibility.

Unlimited* crash avoidance instances - $299.99/mo

10 crash avoidance instances - $199.99/mo

5 crash avoidance instances - $99.99/mo

*crash avoidance may be limited during peak hours and times of congestion. After 12 crash avoidance instances, feature may be disabled without notice due to abuse of the system. All sales are final and minimum 5 year contract required. Price may increase at any time without notice

At the hospital after you crash: dammit I forgot to set up autopay

2 more...

I've always thought that the Tesla craze would fizzle as major car brands start investing in EVs and self driving tech. I'll take a Toyota, Volvo, Honda or BMW over a Tesla anytime.

Sadly Toyota is struggling to make a decent EV years after leading in hybrids. BMW on the other hand has insane efficiency

Still bizarre to me that Toyota had such a lead with hybrids and then went in on hydrogen and missed the boat on evs

They've made some incredible solid state battery leaps recently. For Japan hydrogen actually makes way more sense than a pure EV play too due to the way their grid(s) is set up and their power generation capacity. It's their home market too so they are going to prioritise that.

The solid state press releases Toyota releases every once in a while do look incredible. It would be nice for them to demonstrate and deliver it though because I hope it's not vaporware.

From what I've heard, they have a history of letting other companies trailblaze, and then they come in to refine and perfect tech afterward. They recently tried to be the trailblazer and bet on a losing tech. Hopefully now they are refining the ev game and will come out with something above and beyond what we've seen so far as per their old MO.

My understanding was that the challenge in making the next leap in self driving was not based in hardware (detecting objects with cameras vs LiDAR), but in software. As in, it isn't as difficult to detect the presence of objects as it is to make consistent and safe decisions based on that information.

But using LIDAR, you increase your data's accuracy and dimensionality, giving you more options to play with. It probably won't be a game changer, but it may be better than a camera only system.

Gathering more data, and being able to process it seems obvious as a way forward. How much better is this "new" LIDAR?

Edit: seems Tesla cars doesn't even use LIDAR...

They did. And every other competitor does. Musk believes since humans can drive with only two eyes that cars should be able to as well. Maybe someday, but nowhere in the near future. Cameras miss too much and are easily blinded.

It's also really stupid because the idea is to create a system that's better than humans. And let me tell you, people miss stuff all the time when driving. Tons and tons of accidents are caused by "negligent" drivers who looked both ways and missed someone due to a visual processing error or literally not being able to see something.

That's not necessarily true. What you get is two separate things inputting raw data into a system that both need to be parsed. Sometimes, one won't agree with the other and can cause issues with how the car thinks it should respond.

Nobody has a fully working system at this point, so it's premature to make claims about what hardware is and isn't needed. It may very well be that LIDAR is a requirement, but until somebody figures it out, we're all just speculating.

You can, today, download an app and go ride in a self-driving car around multiple US cities. All of those cars use LIDARs. Sensor disagreement is not a major issue because sensor fusion is a very well-understood topic.

Yes but they geofenced those cars into areas with the most optimal conditions for autonomous driving. What happens when you take the car on the freeway, a suburban neighborhood, or a mountain pass?

Self driving cars are great and all, but can we get someone seriously working on alternative fuels? EV is really pretty unsustainable. All the resources going to build batteries that are unrecycleable is a massive waste in my opinion. And the unless something drastic changes, the ranges that are needed for logistics and America aren't going to ultimately fix anything.

If they can create an alternative fuel that is significantly less polluting, or figure out how to make hydrogen less explody, the existing infrastructure worldwide of gas stations can still be efficiently used. And hopefully there will be a to retrofit existing vehicles to use this alternatives.

batteries that are unrecycleable

Is this actually the case?

My understanding is that EV batteries are actually very recyclable, up to 90%. I imagine it's more labor intensive than your conventional lead-acid batteries though.

I mean, shouldn't we be working on both? Just because they're working on one, that doesn't necessarily mean they're not working on the other.

They also are working on alternative fuels in a big way. Japan have made some incredible leaps with hydrogen/ammonia based production and fuels, and solid state batteries are looking to be pretty game changing. The EU also included a huge budget to invest in green fuels research (likely because of automotive companies lobbying for it) so plenty is being done. Even if EVs aren't the best currently, increasing the size of the market for them will continue to create investments in serving those markets more efficiently, so we absolutely should keep investing in both.

I'm not an expert or anything, but I doubt we'll see a price-competitive synthetic fuel in the time it takes for renewables to become the standard. Renewable electricity gets cheaper as more panels and turbines are built, so it makes some economical sense too.

I think we should NOT allow self driving cars.

> but but muh autopilot in planes

No! Planes are inherently safer to fly than it is to drive cars. Planes have much more room and degree of freedom to maneuver. They are also monitored by air traffic control and the pilots are supposed to be highly trained and fit. Cars are restricted to one plane (heh) and any two bit yokel with stroke-diabeetus-fetal alcohol syndrome from Bumfuck Florida can get a license to drive a car. You can't pull up or dive in a car more than once.

> buh buh buh make every car automated

You just reinvented the train.

My favorite part of a train is the part where it drops you off right at your door.