Volvo Will Sell LIDAR-Equipped Self-Driving Cars To Customers By 2022 (theverge.com) 105
Volve says it'll start selling LIDAR-equipped cars in 2022, which the Swedish automaker says will be able to drive themselves on highways with no human intervention. The Verge reports: [M]ost LIDAR are ridiculously expensive, with the leading suppliers pricing theirs at around $75,000. But now, Volvo says it has found a LIDAR maker that can produce the sensors cheap enough to justify installing them on its consumer vehicles -- which it says will allow these cars to drive themselves. In 2018, Volvo made a "strategic investment" in a little-known Florida-based LIDAR company called Luminar to use the startup's high-resolution long-range sensor to build self-driving cars.
It's an ambitious plan that carries its own risks and sets Volvo apart from its competitors, many of which are planning to launch self-driving technology as part of fleets of robotaxis rather than production cars for personal ownership. They argue this will help amortize the costs of not just the LIDAR, but also the high-powered computing power needed to enable self-driving cars. But Volvo believes that by limiting the operational domain -- or conditions under which the car can drive autonomously -- to just highways, it is creating vehicle technology that is not only safer, but less costly as well. Volvo says it will roll out its self-driving highway feature, dubbed "Highway Pilot," as part of its next big platform update, the Scalable Product Architecture (SPA2), which will arrive with the next-generation XC90 SUV in 2022. SPA2 will also underpin the automaker's upcoming electric vehicles, the Polestar 3 SUV and the XC40 Recharge.
It's an ambitious plan that carries its own risks and sets Volvo apart from its competitors, many of which are planning to launch self-driving technology as part of fleets of robotaxis rather than production cars for personal ownership. They argue this will help amortize the costs of not just the LIDAR, but also the high-powered computing power needed to enable self-driving cars. But Volvo believes that by limiting the operational domain -- or conditions under which the car can drive autonomously -- to just highways, it is creating vehicle technology that is not only safer, but less costly as well. Volvo says it will roll out its self-driving highway feature, dubbed "Highway Pilot," as part of its next big platform update, the Scalable Product Architecture (SPA2), which will arrive with the next-generation XC90 SUV in 2022. SPA2 will also underpin the automaker's upcoming electric vehicles, the Polestar 3 SUV and the XC40 Recharge.
I can also make empty promises (Score:1)
In 2022, you will also be able to purchase a droid which will take care of all your daily needs. It also gives excellent hand jobs.
Re: (Score:1)
This is what happens when you hire community college dropouts to "edit" a once-great website. Seriously, look at BeauHD's bio. And look at his photo in mobile view. He's like 18.
Re: (Score:1)
Robotaxis? (Score:1)
I imagine robotaxis are practically dead in the era of covid, unless there's a way to sanitize the car between riders.
Re: (Score:3)
Fairly easy if you design the interior for easy sanitization with built-in IV lamps (and a damn safe control system to ensure that the car is free of living things and the blinds are down and unbroken).
Re: (Score:3)
*UV lamps, dammit
Re: (Score:2)
*UV lamps, dammit
IR can also work by heating surfaces. Covid is deactivated by heat at about 60C (140F).
Re: (Score:2)
*UV lamps, dammit
IR can also work by heating surfaces. Covid is deactivated by heat at about 60C (140F).
further detail the economics will be the problem
COVID19 within 30 minutes when heated to 167 degrees Fahrenheit. According to ConsumerLab.com, it took 60 minutes at 153 degrees and 90 minutes at 132 degrees to deactivate the virus. At 99 degrees or lower, the virus remained quite infectious for two hours
that leaves a vehicle that is unusable for sometime afterwards extremely hot when you get in and can you guarantee every spot has been heated not to mention the cost of doing it and then possibly cooling
Re: (Score:1)
*UV lamps, dammit
IR can also work by heating surfaces. Covid is deactivated by heat at about 60C (140F).
So it's time to combine cars with saunas?
Re: (Score:2)
Fill the interior with ozone, much more effective, and fast acting!
Do make sure nobody is on board!
Re: (Score:1)
As any taxi driver will tell you, viruses are the least of your cleaning worries.
Even having an internal camera won't catch things like people spilling stuff in the back seat where the front seats obscure what is happening, or worse wetting themselves after a night on the town.
Re:Robotaxis? (Score:4, Insightful)
I imagine robotaxis are practically dead in the era of covid, unless there's a way to sanitize the car between riders.
I imagine that it is a lot easier to sanitize an empty car than to sanitize a driver.
Re: (Score:2)
not when the driver is in an cage like Death Proof
Re:Robotaxis? (Score:4, Interesting)
I imagine robotaxis are practically dead in the era of covid, unless there's a way to sanitize the car between riders.
I imagine that it is a lot easier to sanitize an empty car than to sanitize a driver.
Like those automated public toilets on the streets of Paris, the cars could auto-sterilize their own interiors between fares.
Re: (Score:2)
Where are they going to get a supply of water to wash down the interior of the car with?
And even those toilets get periodic inspections and cleaning by a human.
Re: (Score:2)
Re: (Score:1)
Why sanitize the cars when you can just sanitize the passengers. :)
As soon as the doors are closed, hose em' down with Purell.
Re: (Score:2)
I started to make a joke about white phosphorus, but realized there was a very real danger that someone would be stupid enough to try it. :-(
Re: (Score:1)
Re: (Score:1)
Re: Sanitation (Score:2)
Spartan, non-porous interiors, spray manifolds for gaseous disinfectant, and powerful compressed air drying/ventilation systems could do the job.
Re: (Score:1)
Re: (Score:2)
Re: (Score:2)
Rather the opposite I'd think. Mass transit or taxis with a human driver put people in close proximity with strangers. Robotaxis don't.
Too late (Score:3, Interesting)
Re:Too late (Score:5, Insightful)
Re: (Score:2)
And unlike people, a self-driving car can have perfect knowledge of the stationary features of the roadway because it has a perfect collective memory shared by thousands of other cars (likely including dedicated mapping units with expensive lidar). So it really just needs to see well enough to recognize where it is, and avoid dynamic obstacles. I wouldn't rule out that, say, cameras plus radar (as used e.g. for adaptive cruise control) can do that.
Re: (Score:1)
Are you suggesting that cars share information?
Where eventually the data is stored in a server where it can be analysed to determine peoples movement?
Re: (Score:2)
Are you suggesting that your movements are currently allowed to be secret, if corporations or governments so choose? How do you manage that. We know for a fact the government collects and uses that sort of information to harass people (and corporations and nation states). When Google does the same thing to the public we call it an advertising market. This isn't a change from the status quo, it's just opening the information up to the public, for use in promoting the public good, instead of keeping this alre
Re: (Score:2)
Are you suggesting that they don't already?
Where do you think all the training data that Tesla uses for it's neural net comes from?
Re: (Score:1)
the downside is someone could try to hack the system and cause an accident. For autonomou
Re: (Score:2)
Are you suggesting that cars share information?
He's not suggesting it. It's a fact. Musk and team have talked hundreds of times about how the Tesla neural network works. It's how they train the AI and fleet. Your car is constantly uploading information to Tesla about its surroundings.
Re: Too late (Score:2)
Lidar needs good weather too, in principle scanning/imaging mm wave radar could get through fog though.
Re: (Score:2)
Binocular vision + CNN models can greatly improve accuracy, but there's one thing you can't get around. Cameras require ideal lighting conditions and good weather.
I'm just going to point out that the cameras are actually better than humans at night. The human eye can't consciously control the exposure of the eye while the Tesla cameras can. There is a guy on Youtube that hacked his Tesla at one point and tapped into autopilot to capture what the car was thinking. These are 2 videos shot when the guy was driving at night in a storm. Video 1 [youtube.com] Video 2 [youtube.com]. You'll notice that the B-pillar cameras have their exposure turned up way high to see the scene when its dark and turn i
Re: (Score:2)
Tesla don't even use binocular cameras. They rely on AI to guess depth. It's as janky as it sounds.
Tesla seem to think that driving isn't all that hard. If they can just detect enough of the world to avoid hitting things and follow the lane markings they will be okay. In practice I think they are finding that there are so many corner cases where that fails their full self driving tech is never going to actually work.
Musk is still predicting self-driving Tesla robotaxis this year though. Boundless optimism I
Re: (Score:2)
Thesla uses 8 cameras, 1 radar and 12 sonars.
Re: (Score:3)
Yeah, how have people been driving for 100 years with only a stereoscopic system of two cameras with a ~135 degree field of view on a slow gimbal mount with rotation restricted to ~70 degrees?
You know that autonomous driving can do exactly what humans do if you encounter weather severe enough to impair driving, right? Pull over and stop.
Hey look, I just solved your problem with the exact same solution people have used for 100 years.
Re: (Score:2)
People hit pedestrians all the time in the dark.
Also our gimbal aka our eyeballs are very fast and gives us 8k+ in the narrow FOV that we're currently scanning.
Re: (Score:2)
Polarized lenses are a thing. It's not hard to block glare, and if you have multiple cameras you can polarize for multiple planes and then composite.
Re: (Score:2)
You don't *need* multiple types of sensors. People drive using a pair of eyeballs. It's nice to have multiple types of sensors so you can do better than people.
Conventional LIDAR is still probably dead though. You can build a LIDAR sensor pretty cheap, but it still relies on mechanical scanning, which is slow, clunky and error prone. Time of flight CCDs exist, aren't any more expensive to produce than regular CCDs, and offer LIDAR-like depth sensing on every pixel, every frame.
LIDAR might also have problems
Re: (Score:2)
You use your sense of smell to drive a lot hey? I did have a friend who used to monitor engine overheating by the burning smell, but most cars, which also have multiple sensors, have better ways of doing that.
We're not talking about what's easiest to do, we're talking about what's required. A person can drive with two eyes. Actually, a person can drive with just one. A single camera is therefore sufficient.
As for making it easier, I don't really think LIDAR is the thing. Stereo cameras are superior to LIDAR
Re:Too late (Score:5, Informative)
Not sure I agree.... Tesla's claim of being able to use their cameras and radar as a LIDAR substitute seems to be more theoretical than something they're proving in execution.
I'm a Tesla owner myself, but I feel like they're really struggling at this "full self driving" they're chasing after. Just watch some of the YouTube videos of people trying out the latest functionality where the car can stop for red signals and stop signs automatically. It's pretty awful. The car completely misses some stop signs and occasionally stops at intersections where there's no sign or signal and it has right of way. I know it's just "beta", but they spent a lot of time on this as the only really new thing the car is ready to demonstrate with the latest "AP 3" computer for the autopilot installed.
It still has an overall buggy/glitchy feel to a lot of the autopilot functionality, including the "phantom braking" issues where it slams on the brakes for no sensible reason while it's driving on the interstate. (Sometimes seems to be caused by an overpass the camera mistakes for an obstacle across the lanes. Other times might be GPS drift mistakenly telling the system you're on the access road adjacent to the highway instead of on it?) Also seems like various software updates they push change steering behaviors for better, for worse, for better again... back and forth. You never know what to expect with things like sharp turns it goes into at relatively high speeds.
Re: (Score:1)
And how exactly is LIDAR going to read a speed limit sign or a red light?
Re: (Score:2)
LIDAR should be able to tell the difference between a sign with empty space around it versus a picture of a sign on the back of a truck. An associated camera can read the contents of the sign, gated by the LIDAR data.
I've been living close to where Google's self-driving fleet begins and ends each test day, and I've been tempted to put a traffic sign on the back of my car. Deep learning systems go out their way to make recognition scale invariant, so even a stop sign on a bumper sticker might be an interesti
Re: (Score:2)
Two cameras separated by a few inches could also tell the difference quite easily.
Re: Too late (Score:2)
Re: (Score:2)
Re: (Score:3)
No, I'm pretty sure he's saying that the GGP is misrepresenting what the current intention and capability of the software is.
It's not meant to drive through green lights without you, the attentive driver, telling it to. It literally says it's going to stop at every signal and stop sign it sees unless you tap the autopilot stalk to tell it to proceed.
If you turn this on and start yelling WTF when it stops at green lights, you're just an idiot that doesn't read clearly written warnings when turning on early
Re: (Score:1)
Re: (Score:2)
The car completely misses some stop signs and occasionally stops at intersections where there's no sign or signal and it has right of way.
And how exactly is LIDAR suppose to be able to solve these issues? If the car is missing a stop sign then its probably because the car can't see the stop sign for some reason and if that's the case then LIDAR isn't going to auto-magically see it as well. LIDAR can only see shapes, no color information. LIDAR when paired with vision might help hint to models that a sign might be present at a location if the sign is obscured partially, but there isn't a guarantee that LIDAR would fix these same problems that
Re: (Score:2)
seems to be more theoretical than something they're proving in execution.
Hardly. Before you claim that you need to prove that the examples of incidents were to failure of the camera and algorithms and at the same time prove that LIDAR would have prevented them.
There are several such examples (e.g. the truck going across the highway, and the steering into the divider) which were proven to be nothing at all to do with the sensor logic and everything to do with how the control system reacted to (or ignored) portions of the available information.
Give me a LIDAR and I too can produce
Re: (Score:3, Interesting)
Re:Too late (Score:5, Interesting)
It's a legacy problem. They were expecting to do full self driving within a year of releasing their old 2.0 autopilot hardware. They then did V2.5 and are now on V3.0.
If they had managed to make it work back in 2017 it would have been a big leap ahead. Lidar was too big and expensive and power hungry back then. But they failed and will doubtless fail again this year and by the time they get there, if they get there, lidar will be here anyway.
The problem they have is that they started selling "full self driving" back in 2016 so now they have over a million vehicles with camera-only hardware that they need to support. Not just for customers who paid for it already, but to support their business plan of getting income from robotaxis and selling the technology to other manufacturers. So they can't just switch to lidar now.
Re:Too late (Score:4, Informative)
It's a legacy problem. [snip] so now they have over a million vehicles with camera-only hardware that they need to support
Musk has stated numerous times that if they are wrong and require different hardware that anyone who pays for FSD will get upgraded for free. In fact they've already done so by replacing v2 computers with v3 when people take their FSD licensed cars in for any type of service.
Re: (Score:3)
Yes, but say they needed lidar. How much would that cost to retrofit? Can't just swap out an existing camera and re-use the same wiring harness. It will be a major modification to the car, and probably affect the aerodynamics and thus range.
Worse still Tesla seems to have been relying on getting income from the robotaxi scheme with potentially a million cars already on the road, but if most of them can't do it without an expensive upgrade...
The clock is ticking for them too. If other manufacturers release s
Re: (Score:2)
The clock is ticking for them too. If other manufacturers release self driving taxis or personal cars in the next few years people who bought Tesla full self driving back in 2016 are going to want to know why they don't have it.
I mean the worst that would happen would be a class action lawsuit and maybe they have to spend the money giving out new vehicles, but this would come down to the wording/promises of the package on the site back in 2016 when these people bought the car and package.
I'm not too worried about other companies beating Tesla to market though. The real question about self-driving is whether the solution to the problem space is a fundamentals issue or a data issue. If its a fundamentals issue, then there is some co
Re: (Score:2)
Cameras have their weakness, LIDAR is better but perhaps not perfect (it cannot tell the difference between a brick and an empty KFC box on the road, best to just avoid both.)
I would avoid the KFC box, too. I don't know it's actually empty, and it might not just be biscuits in it.
However would you want to buy a sound system that used AI or special filters to make good sound from crap speakers or just buy actual good speakers then add the enhancements on.
If the cheap speakers with AI & filters sound as good as the good speakers at half the cost, I'd probably go with that.
What's the cost of killing someone? (Score:2)
UBER tried something like this and killed someone in the process. On an automated vehicle, the more redundancy you have and the more sensors the car has the less likely it is that it'll encounter conditions that'll end up killing someone. Cameras don't work well in the dark.... Lidar works amazingly well in the dark.
Re: (Score:2)
In the accident, Uber turned off the Volvo's Automatic Emergency Braking system. Existing radar based technology could of prevented the accident. Lidar not needed to prevent this accident.
Re: (Score:2)
UBER tried something like this and killed someone in the process.
This was a human error. Someone disabled the emergency braking system on that car by mistake.
Re: (Score:2)
They had also switched off the LIDAR
Re: (Score:2)
Volvo currently sell a plug-in EV which is about twice the price of competing models. How they're selling any is a mystery to me, but it seems volvo owners are happy to sell out for their products.
As much as my "liedar" is going off right now, I wish them luck with this. It's time Tesla has some proper competition, and some that wasn't very much US focussed would help the rest of the world to boot. I suspect LIDAR cars will be sold in 2022, with some highway self-driving, but actual driving around towns and
Re: (Score:2)
It's not that difficult to see that nearly every Tesla autopilot crash was due to inadequate sensor data. Their software is good, probably the best, but without Lidar no amount of software is going to make up for what it can't see.
Re: (Score:2)
It's not that difficult to see that nearly every Tesla autopilot crash was due to inadequate sensor data. Their software is good, probably the best, but without Lidar no amount of software is going to make up for what it can't see.
And yet the IHS data already shows autopilot is far safer than humans.
Re: (Score:2)
https://www.youtube.com/watch?... [youtube.com]
Is that what you call LIDAR accuracy?
Why use LIDAR? (Score:1)
Simple stereo vision will determine distances pretty well.
I drove a Volvo on auto down a motorway last year -- it did a pretty good job. But suburban traffic is a different matter, and I do not think that Lidar is relevant to the solution.
Re: (Score:2)
Drive around on a moonless night in the countryside with no lights.... see how your volvo goes then.
Re: Why use LIDAR? (Score:2)
Nothing a some IR floodlighting won't take care of.
The only thing LIDAR has over stereo is energy efficiency.
Re: (Score:2)
Tesla may be able to make a car that can operate fully autonomously without
Re: (Score:1)
Re: (Score:2)
Hundreds of vehicles all using lidar at once (Score:2)
I'm wondering how these lidar systems will function when there are hundreds of them on the road all at once.
Re: (Score:2)
The speed of light is damn fast. A light source would have to be fiendishly well-timed to interfere with a LIDAR return. Each LIDAR device could shift the laser pulse timing pseudo-randomly to avoid interference.
Re: (Score:2)
Pulse modulate. Instead of sending an on/off you can send:
On OFF OFF vs OFF ON ON vs ON ON OFF vs ON ON ON etc.
The odds of interference is already very very low spatially. You could increase the chances beyond the number of cars possible to fit into a space.
Re: Hundreds of vehicles all using lidar at once (Score:2)
They will probably have to start using cdma, dunno if the existing systems already do.
No they won't (Score:2)
The chance that anyone's going to sell me a car that can self-drive around London (without leaving a trail of destruction behind it) in 2022 is zero.
Mad Max (Score:2)
Yes Cheap Sensors Are keeping Me Safe (Score:1)
Demolition Man got it right again (Score:2)
Re: (Score:2)
So what you're saying is, I need to buy stock in Taco Ball.
(J/K, I already invest in PepsiCo, Inc)
Armchair warriors much? (Score:3)
It's funny to me how many people on here who have never driven a Telsa seem to know so much about what they have (or haven't) accomplished with FSD.
My advice. Shut your trap. Go spend a day driving one (or letting it drive you). Come back and tell us little you actually know. I promise to accept your apology without an "I told you so". ;-)
I won't care until 2022 (Score:2)
This is still emerging tech and given the state of the global economy now (pandemic recession) and the underpinnings of the success of the last decade (over-investment in unprofitable unicorn startups), I'm not entirely convinced that even the more stable companies will be able to produce lawyer-safe level 5 autonomous vehicles affordable to any significant portion of the American population.
1. The tech is hard & expensive. Those trying to corner this market early are following the "get big quick" metho
Re: (Score:2)
1. The tech is hard & expensive. Those trying to corner this market early are following the "get big quick" methodology which means they can fail once enough people call "BS" on their finances.
Agreed. But Tesla isn't one of these. They are profitable, and all indications show that will only increase.
2. The liability of it all is a HUGE pill to swallow. The company responsible for the tech is the "driver" and thus liable for all errors on the road. Currently, liability is distributed to each human driver and the recuperation of losses is limited by what the driver can pay. Centralizing that liability on a few large corporation ensures financial ruin unless the vehicles are nearly flawless.
The solution for this already exists. Some states have something called "No Fault" insurance. Where everyone just pays for their own no matter who is at fault.
3. Cars last a long time with minimal maintenance today. People are losing money fast right now, so it's not as if they'd have the disposable income to swap out for a new AV if one were to hit the market.
Not everyone will buy one immediately, not unlike any other upgrade.
Re: (Score:2)
Freakin' great response, friend. I shall retort!
1. Tesla is recently profitable as an entire company. That's correct and even as a Tesla doubter, I was happy to see it. However, as someone who works in sustainability and who watching the markets (especially for financial BS), I've never felt secure in discussing Tesla's financials on the whole because they intertwine their R&D, auto mfg, battery mfg, and solar business. A massive portion of Telsa revenue is sustainability subsidies and the sale of emiss
Re: (Score:2)
I'm not entirely convinced that even the more stable companies will be able to produce lawyer-safe level 5 autonomous vehicles affordable to any significant portion of the American population.
Level 5 is stupid. Nobody will focus on Level 5. You don't need a car that can drive to a lake house down a dirt road. Cars will focus on the areas with customer: suburbs, cities and highways.
Even Level 3 like what Volvo is pursuing has enormous value. Fly or lay down in a sleeper bed and wake up in a new city with your car the next morning? $40 in electricity can get you and a few of your friends 1,000 miles away. That's valuable.
Level 4 is also very useful. If 95% of the roads in the country ca
LOL no they won't (Score:2)
Bleatings from SDC fanboys about human drivers being dangerous are irrelevant. If your shitty AI can't be better overall in ALL situations than a human driver
Always ten years off (Score:2)
This (self-driving) will be ten years off for quite a while, outside of some specific well-controlled situations.
Better sensors are not the answer. Nor are faster processors or more clever algorithms. The fact is that human wetware is still in a league of its own when compared to any artificial "intelligence" either in existence or on the drawing board.
The human eyeball can be outdone by various lenses and sensors, but it's our nervous system's ability to interpret the input that leaves digital systems in t
Re: (Score:2)
This (self-driving) will be ten years off for quite a while, outside of some specific well-controlled situations.
I take it you've never been driven by a tesla before? I took mine on a 200 mile road trip just recently. I drove maybe 5 of those miles. They car drove the rest.
Building a cheap LIDAR is 0.001% of the problem. (Score:2)
LIDAR gives you range - and not much else.
The software to extract range from stereo cameras is pretty routine stuff these days - and with cameras that are moving at known speed and direction, it's even easier.
Since you can buy a whole bunch of cameras for $10 - I don't see the joy of LIDAR.