Lidar vs. Tesla: the race for fully self driving cars
HTML-код
- Опубликовано: 27 июн 2023
- In the world of autonomous vehicles, lidar sensors are the center of debate. Self-driving car companies, like Cruise and Waymo, use lidar as the key ingredient to advance their autonomous vehicle navigation while skeptics, like Elon Musk, claim it to be useless. Transportation editor, Andrew Hawkins, explores the landscape of both and why exactly lidar continues to be at the forefront of complete autonomous driving. #technology #cars #lidar
Subscribe: goo.gl/G5RXGs
Like The Verge on Facebook: goo.gl/2P1aGc
Follow on Twitter: goo.gl/XTWX61
Follow on Instagram: goo.gl/7ZeLvX
Follow on TikTok: bit.ly/42VeCVU
The Vergecast Podcast: pod.link/430333725
Decoder with Nilay Patel: apple.co/3v29nDc
More about our podcasts: www.theverge.com/podcasts
Read More: www.theverge.com
Community guidelines: bit.ly/2D0hlAv
Wallpapers from The Verge: bit.ly/2xQXYJr
Subscribe to The Vergecast on RUclips, new episodes on Wednesday and Friday: bit.ly/3I6nJtz 
Наука
How ”self-driving” do you want your self-driving car to be?
Like Bumblebee
People like me definitely need self driving cars. Every single car should be equipped with a self driving option in the future. This is the only way I could ever own a car, or else I will be riding on cabs forever.
Autopilot and Full Self Driving are two different systems and as a reporter you should know that by now 🙈. You should Compare Tesla Full Self Driving crash data to that if Waymo and Cruise along with the number of Miles traveled. Tesla autopilot it's free while Full Self Driving cost $15k so the cost of Waymo and Cruise lidar alone covers the cost of software and hardware of Tesla Full Self Driving technology.
that is the question that is trying to be answered by all the auto manufactures now. Tesla's answer is "absolute", they want the car to be able to have a destination point entered, then get there fully on it's own with zero input from the driver. other companies (GM, Ford) are looking at just removing the "tedious" portions of a drive, such as long highway commutes in heavy traffic, not point-to-point service. personally, i'm with the latter group, i don't need "self driving" to get me 6 blocks to the grocery store. however, having something that will effectively "drive" through bumper to bumper rush hour traffic, yes please.
Easy -- all the way (without compromising safety, of course).
Over the last 10 years I have witnessed both my parents having to give up driving through age-related ill-health. Even though they lived in pretty much the perfect location for walking access to their doctor, dentist, high street shops, church, etc. losing the car was still a major blow to their lifestyle. Then you have all those elderly people who continue to drive but shouldn't.
So, I'm hoping that by the time I reach the age when I should no longer be behind the wheel, self-driving cars are a reality. For the elderly, disabled, and infirm alone, this would be an amazing technology to have.
If we are going to go with self driving cars, they need to see better than we do.
Well, they don't blink, have cameras on every side, have good resolution throughout the entire field of view... How's that?
@ It's a good start. Now how solid is the protection in preventing them from controlling where we are allowed to go?
They already do see far better than we do. Can you see everything in 360° and analyse everything that's happening instantly?
We currently don't have self driving cars due to the brain, the vision side needs no update. Humans only wreck because we have an attention problem.
They need to be better at processing information. Not seing better. Reaction time is probably the higher cause of accidents compared to visibility
My understanding was cars like Waymo only operate in pre-mapped environments, where Teslas are built to “read the room” and figure out what’s in front of them. LiDAR or not, that approach makes a huge difference in the performance and limitations of the autonomous systems.
For the longest time some tech uses lidar maps, which doesn't allow it to adopt to roadways that's not pre-mapped. No mention of this in the video is a huge omission.
Geofenced companies do it not because they can only operate in pre-mapped areas. They do just as much of "reading the room" if not more (due to having more sensors in both quantity and variety). Even in the geofenced areas, roads change every single day, from constructions, pedestrians, double parked cars, etc., pre-mapping alone would be insufficient to begin with. Geofenced companies can lift that restriction if they want to, but they are in a world of risk if they do driverless without a geofence right now. It's the same reason why Tesla still requires a driver when using FSD.
Elon just tells you a new lie everytime you need to hear one.
I know about 10 people who invested in Tesla because "everytime someone drive a Tesla, the resolution of the map increases". Now, it's the opposite.
It is very clear to me that Tesla has already lost the race for autonomous vehicles. Dispite all the money thrown at them! it's quite impressive..
Yes, the difference is that while it should theoretically generalize better than mapping, Tesla's approach is enormously worse at actually driving.
Im not a musk fan and I think his choice to remove radar was wrong. BUT were you unable or unwilling to compare equal data sets between Waymo and telsa? Could you not find Telsa data per one million miles or did you not even try? Just seems like bad reporting to put to compare two completely different stats.
Super quick google and I found "Autopilot got in an accident once every 4.31 million miles". So, did you not even bother to google it? Did you find different data?
I'd love an explanation because these seems like supremely shotty reporting from The Verge.
@@LiamMarcon And the data in the article includes traffic aware cruise control accidents.
They’re adding HD Radar into the new Teslas. This will complement the current sensors rather than compete with them.
Also Waymo and Cruise don't cruise on highways, do they?
Yeah this was clearly not unbiased reporting. Just another example of the media going after Musk. He appears to be Trump 2.0 in that regard.
Removing LiDAR was about cutting costs at the risk of accidents that could otherwise be avoided. 2d cameras are not great at detecting depth especially in dark environments where the sensor gain needs to be increased to the point where the image becomes noisey and far less useful. Edit: Please read my below comments for more detail on this if you are skeptical.
Spitting more facts than the video
@@VincentKarabouladMusique I was kidding. I made a comment earlier about using cameras to detect objects and light instead of LiDAR
If humans can do it, why can’t cameras and computers?
@@chi7891because cameras are not humans? This whole comparison is silly, technology and humans both excel at certain things but the human eye in particular is incredibly hard to recreate in terms of clarity, adaptive light sensitivity; etc. again, camera sensors even today struggle in low light without heavy noise reduction/processing. This processing/noise reduction is nice for looking at images but isn’t super useful for autonomous vehicles because it ultimately either creates artifacts or just smears detail depending on the technique they use. The solution is either LiDAR or much larger camera sensors, both would be more expensive than what Tesla is doing right now.
@@Kdkjdjewerdnxa the roads are designed for vision. If camera improvements are needed, that’s fine, but you won’t be able to recognize things like stop signs or traffic lights with LiDAR.
With my commute times, and no train options, a driverless vehicle would get me back 20 hours a week in at least reading or working on my hobbies
It doesn't seem fair to me to compare Waymo's accident record in miles and Tesla's in years. It would have to be in the same unit (distance or time), although distance seems to be the best indicator.
or at least get the Tesla crash data for the specific locations (Phoenix metro area and SF metro area). There are several reasons why those two areas are chose for this type of self-driving testing.
@@rotinoma’s not so much the location but the number of miles. It simply doesn’t make sense to compare unless he had divided the number of crashes by the number of miles
That was a very cheeky biased comparison honestly
We don't need self driving cars. We need self driving buses.
Imagine new york streets full with only long autonumes buses with 50 people each - it'll make traffic much more bearable
Or think in reverse and make less destinations for people to go to so that large buses make even more sense. Mega-City One here we gooooo
CYCLING
Longer term, you can imagine a combination of both types of vehicle -- communal and individual -- forming "virtual trains" as the travel down the major routes together, all coordinating with each other and communicating with the city grid system which optimizes the traffic flow to get everyone to their destinations as efficiently as possible. The density and speed of the traffic would be much greater than possible with people behind the wheel.
Perhaps the HOV lanes on many highways will be the first to be converted to self-driving only...
So that when it crashes, it takes out people in bulk👍
@@EnglishMike This would never work. It's tech bro nonsense. It only takes 1 car that isn't self-driving to disrupt the whole flow.
What happens when one of the cars gets hacked? You could melt down an entire city from across the globe. You could even shut down an entire country all at the same time.
Self-driving cars already exist - they're called trains.
Its weird that Lidar cars can basically only drive in tiny pre-scanned areas, but Tesla with only cameras drive everywhere in North America. Even with Waymos avoiding freeways and driving the same routes 1/3 as fast as Tesla, its still a good company, but cant scale fast at all.
1. Tesla's can't drive anywhere on their own.
2. LIDAR can be used outside of a geofence. Mercedes uses LIDAR which allows them to reach level 3 for high traffic scenarios under 40mph.
Lastly, Waymo actually can take highways as long as a safety driver is in the car. I think they're less willing to risk a high speed accident until they're fully confident that it can handle it 100%
@@TheSpartan3669 Mercedes is also geo fenced and not a single customer got the drive pilot yet xD
@@TheSpartan3669 Tesla is trying to solve a much bigger problem. and they need data to do it. The fact it has millions of cars on the road with many drivers actually paying to Test FSD for Tesla is a huge advantage. The challenge isn't tiny areas of America avoiding the more complex scenarios. It is to drive from point A to point B in an efficient and safe way.
Waymo has the safe part in very select areas, Tesla has the point A to point B. To me Tesla will grow safer far quicker than Waymo can extend it's range of travel.
Well, Teslas CAN'T actually self-drive ANYWHERE, so any theoretical scaling doesn't matter.
@@Blaze6108 They can self-drive in myriad locations, but just not BY themselves (read “alone”) and without supervision. No car you can buy can yet. No one else is as close to that goal as Tesla, but china has a few that are pretty close (but still need pre-mapping which Tesla doesn’t).
I miss Verge science, and we need more content like this ❤. Your content recently has been all over the tech and fashion tech map.
There is one fundamental errors in this video. You compare 2019 Autopilot data (where the cars were NOT driving by themselves in any way, it's basic lane assist and adaptive cruise control) with more recent data from the other companies where the cars are actually driving by themselves. Get some data with the latest FSD beta from Tesla, and see if the result isn't much better (factoring in the number of cars on the road from each company).
The Verge videos are usually very thorough with it's data, but this one is not.
Lol the verge is always a joke.
Waymo has had 2 crashes in the previous 1 million miles? Tesla cars drive collectively 1 million FSD miles per day. That is why the statistic is very skewed, you compare it to 4 years (4x365=1.460). This statistic is presented very misleading in your video.
Exactly
Verge being biased , shocker.
Also he compared it to Autopilot (Traffic-Aware cruise Control) and not FSD (Full Self Driving) - there is a difference
@@kayslay6195 good point, that makes it even worse 😅
Should also compare to human drivers too.
(Without checking) I would guess they are both significantly less likely to have an incident/accident and fatality.
Truly HORRIBLE crash data comparison between Teslas on Autopilot and Waymo cars. Waymos have been in 2 crashes and 18 "minor contact incidents" (whatever that means) versus Teslas have been in 736 crashes since 2019? This isn't an apples to apples comparison. You can easily google "Tesla crash data" and see that on Autopilot, there is an average of 1 crash every 4.8 million miles on Autopilot
Agreed. AWFUL reporting. Truely bad. I can't tell if they're incompetent or bias. Probably both.
P.S I don't even like Musk! But I hate bad reporting more.
@@LiamMarcon on top of what you both said autopilot isn't FSD Beta which is the appropriate comparison until Tesla has a true driverless mode
@@KaceyGreen Exactly! Not even the same things
@@LiamMarcon Haha love the transparency
What about Waymo only operating in the City, where crashes Are much more likely to happen and Teslas putting a big Chunck of the Miles on the Highway, where Even normal level 2 systems will prevent most of all crashes. Again, you also do not compare Apples with apples 😅
I think the Internet of Things is the ultimate way to self-driving because if everything knows about each other, than the algorithms can do the rest, but putting each thing into the IOT is the hardest part of the job.
I used to think that too. But devices are more reliant on each other which can be bad. Having a separate system that can work on its own is more secure.
@@dkdeep5505 i agree, but what im thinking is they dont reliant on each other, they just been noticed that each other is exist and making dynamic adjustments on its on, that would be different way for vehicle to sensing the world
@@adamlin5079 how would they notice?
This is a terrible idea, pedestrians and generally things other than vehicles are not IoT. What are you going to do, make it mandatory to wear some IoT chip or install an IoT app?
Uh first no that is extremely dangerous second impossible first if the internet gives out goodbye car second the internet is filled with hackers you are as good as dead if they decide to use the internet to fully control the car.
Hey guys i have a prjoect where i have to come come up with a way to stop damages due to close reflective objects cuasing the incoming pulse to damage the system. Tbh i didn't even know this stuff existed till a few days ago, im in second year of electronics..any suggestions??
I think I imagined that autonomous vehicles would share information with each other (just those within a certain distance, or on the path you your on). So it would be kinda be like a omnipresent system via cameras.
I did originally think using a combination of lidar and cameras were better (for Musk's example of a plastic bag floating in the way, causing a lidar system hitting the breaks), but now, I realise technology is getting more and more interconnected.
I think that will happen in china faster than in the west and the rest of the world , then again what are you going to do about the rural areas? that can work in urban areas but rural areas are literal dustballs territories... the vehicles will still need internal logic...
@PrograError true... I have no idea. maybe just gonna have it work as it is now? 😅 cuz if there is not much traffic, the only thing the car should be worried about is local pedestrians, wildlife, livestock, etc. Which will all be covered AI anyways
Kinda sorta, but it's extremely bad practice to rely on info sharing if you're not in a truly closed system. For example, autonomous subway trains can get away with having barely any sensors at all because the environment is so controlled that info sharing can do all the sensing for them. But for cars operating in civil environments, it is completely insane to rely on anything less than accurate perception.
What was not mentioned ( fatalities) in the Tesla Accidents as to how many million miles driven by tesla , also not mentioned that for lidar only you need to be geofenced and not able to drive all over
Yea the video is misleading. The % of Tesla incidents is lower than Waymo incidents.
@@chrisak49 How much of that Tesla accident data happened while the car was fully autonomous? Driver assist mode is very different from fully autonomous. With an accurate like for like comparison, no accurate conclusions can be drawn.
@@EnglishMike The video is also misleading cause the guy kept saying Autopilot which is just adaptive cruise with lane keep. He doesn't include data for full self driving data because he can't get it and is probably too lazy to dig for the data.
George Hotz was the first to say that we don't need LIDAR for Comma AI. They should have done the courtesy of mentioning it.
I've seen fully autonomous vehicles in SF, Phoenix, and Austin ALL with LiDAR equipped. Tesla's FSD camera system requires a driver. It may change, but thats where we're at in 2023.
What you said is true, but a bit of an oversimplification. Waymo and Cruz are Level 4 autonomous and Tesla is Level 2 today. However Tesla is aiming for Level 5 autonomous, Waymo and Cruz are not, they are just working towards better Level 4. If you ask who is in the lead towards Level 5 its Tesla all the way, no one else has even started.
@@chrisoconnell8432 Cruise, and Waymo are fully autonomous, level 5. Just allowed to run in certain cities. Tesla's full self driving is currently level 4.
@@bingeMAFIA In Level 3, 4 and 5 the manufacturer is liable in an accident, Level 1 and 2 the driver is still liable. So Telsa is Level 2. A very capable Level 2, but still 2. Cruise and Waymo REQUIRE detailed HD maps of the area they will drive. If tomorrow you told them they have permission to drive NY City they couldn't do it because they don't have detailed maps of that city yet. Its just not possible to maintain up-to-date maps of every street on Earth. That is why their system will always be Level 4 without a rewrite from scratch.
@@bingeMAFIA - no, a level 5 car must be capabable of mastering all traffic situations. It takes millions of driven miles to feed the ai and make it drive better than a professional car driver. And guess who's going to have it first.
As someone who worked with GIS, photogrammetry and with reduced acuity in one eye due to surgery ( low stereoscopic vision) I can tell you the following. LiDAR does not only produces a 3D map of the environment but for every 3D point scanned detects the doppler shift so from a single image you can tell what is around you, what is moving and what is standing. Theoretically speaking , stereo cameras or arrays of cameras can achieve near human levels of seeing but your brain can Id each an every part of the environment, can tell what is moving what is dangerous can gauge distances based on proportional sizes and many more. Also cameras are susceptible to a variety of optical illusions (forced perspectives etc) different items with near same colors and so on. I do not know if LiDAR is the cure all solution for Level 5 autonomy ( the holy grail) but is close enough.
I've taken a cruise several times in Austin, and the cars just arent there yet. The cruise takes 25 minutes to take me from our favorite bar to my apartment, which would only be 10 mins in an uber. They are somewhat jumpy in turns and cruise has very clearly only allowed the cars to travel along predetermined routes, and avoid certain maneuvers. This causes the cars to bunch up into little trains in certain parts of downtown. Any way you slice it, we are a long way off from general purpose, level 5 autonomy.
Can lidar read Speed limit or text in signboard?
Autopilot < Waymo < FSD
The accidents that were reported are for Autopilot which is basically Cruise Control + Lane Keep Assist. No doubt it's a misleading name but it has led to the average person believing that Teslas are already fully autonomous which I guess was important branding.
Waymo uses HD maps which means that their cars can only drive in areas that are already mapped. This is expensive and time consuming which is why it's limited to California.
FSD on the other hand relies on vision meaning the system is self contained and given that it learns from the millions of cars world wide will eventually be able to drive themselves without supervision (L4 at least). It will never be as safe as using Lidar but it's a tradeoff that's worth it.
"but it's a tradeoff that's worth it."
Without accurate like-for-like comparison data, we don't even have any idea what the tradeoff is.
IF FSD was better I'd expect FSD robotaxis to be driving around by now. Like Waymo does.
Lidar is more accurate. Especially at night. Yes cameras are okay and will work similar to lidar as long as there is enough light. Programming lidar is easier, and accurate, but expensive. Programming camera to create ML is difficult but justifiable in the long run, and hardware itself is cheap.
No programming lidar also requires programming the system to understand obstacles and do its work. Lidar just provides raw data the same as cameras.
@@hilmyakatsuki1665 If that raw data provides more and higher quality data about the objects around the car, then the software processing that data will be easier to code and be less prone to mistakes (all other factors being equal).
@@EnglishMike again it's not that easy. Otherwise lidar could take over the mainstream market already! Both systems need to be taught how to differentiate between objects. The Lidar system also needs a 2d system to recognize and validate the data before deciding anything. Tesla also taught and validates their 2d camera system with lidar on the testing phase before using that model on the real road for customers without lidar. Yes lidar can help but at this point there's nothing indicating that lidar provides anything meaningful more than 2d cameras.
@@hilmyakatsuki1665 Lidar has been way too expensive to be in mass produced consumer products.
Lidar isn't confused by shadows, bright sunlight or oncoming headlight beams, all of which have to be overcome when using camera systems. If AI processing has reached the point where they can be effectively overcome, then great, but it took a lot of extra work to get to that point.
Lidar is not more accurate. Especially when it's rainy or foggy it can completly fail. And this is the moment where a camera based system has to take over 100 %! Cameras (=eyes) are human proofed. There will never be cars driving with turned off lights (f.e. to be visible to animals or pedestrians) and humans don't make car accidents because they lack a 6th lidar sense. Humans overestimate themself or get distracted. It's always the same actually. So no need to change the sensors. It's the brain that has to be replaced by an AI to make driving safer.
Brought to you by Luminar 😂 So transparent...
Lmfao right?. What an agenda. So poorly argued too.
I love the bit where he tries to explain the usage of LiDar and it starts out as "Well, apple products have it" without any mention of their usage on these devices. Unintentional dunk on apple tbh.
He also didn't talk about the specific usages in farm equipment, warehouse vehicles, and aviation. Was it an unintentional dunk on those applications too? Nah, he already explained what lidar does and how it works at a high level, mentioning Apple and the other use cases is meant to briefly help visualize and associate how this sensor is applied elsewhere. Going into specific applications for products unrelated to this video means going off-topic.
Cars at night are not essentially blind!
1 cars have lights or we wouldnt be able to drive them.
Cameras are also much more sensitive to small amounts of light.
Also auto pilot is not full self driving! Auto pilot is just advanced cruise control.
It seems to me that to be a mostly vision-based autonomous system, you'd need REALLY good cameras. Ones with insane dynamic range and resolution. After all, the same camera has to be able to look straight into a late afternoon sun and also be able to process a poorly lit country road on a moonless night. I think Elon's comments about lidar sound to me like a CEO trying to justify a cost-cutting business decision.
This was an excellent video. Short and sweet.
I think you should have also included stats of accidents with no driving assist at all.
And the stat of 1M miles with Waymo. 35M miles on the Full Self-Driving beta (although I couldn't find any accident data regarding the beta). Also the Tesla stats include traffic aware cruise control, which isn't really self driving at all. It does seems like comparing Apple to Oranges.
Obviously, humans are the problem here. Taking us out of the equation will reduce accidents.
Yeah, that section was an apples to asteroids comparison. Not good.
@DrieStone problem with the Tesla stats is the whole controversy about it disabling itself prior to impact. Just a cursory video of near misses on RUclips or Twitter where the driver stops FSD from Killing a cyclist or running a busy stop sign and uts clear that the FSD safety stats are pretty useless and can't be compared to an autonomous vehicle like waymo where an accident is an accident, black and white.
@@torotech makes sense thanks
Does LiDAR outperform camera data in adverse conditions? That would be the #1 question to me. I thought that it did, but I read otherwise.
it does not. doesn't even work in rain. There has to be someone paying these channels to promote lidar like this cuz idk why they hype it up so much. it clearly will not be a part of fully autonomous vehicles in the future, camera and radar combo is the solution.
It depends on the factor you're interested into. For example non-stereoscopic cameras cannot do 3D vision by definition, so if you like having depth perception LIDAR is infinitely better than them.
when you talk about how many people hurt using lidar while providing the miles, you should as well compare it to tesla while providing the miles, to make them on a comparable scale
He should have also used data from city driving only and not highway... As he talked about autopilot, which is the one you can use only on highways... 2x speed kind of increases chances of injury. Tho at same time, its not even level 3 or 4 or 5, so entire fault is on humans and not on auto pilot, but heck, people will be doing what they do best "Spreading fud for the sake of it"
This is very intressant, and Lidar is needed for full self drive, 2 in front and 1 in rear.. but i will hate if the install then like NIO does on there cars, that will be a deal breaker. also WHY is all camera in front NOT behind the headlight glass.. they will last longer.. and be much better protected.
I would like it. I can't see enough to drive due to an accoident, and I used to love driving.
Lidar doesn't work in the rain or fog so it's pretty much useless in most places
I literally stopped updating my Model 3 in August 2022 because of the update disabling radar. I don't want a sensor I bought, that helps to sometimes see more than me to be disabled. I don't want autopilot to be as handicapped as me by low angled sunlight or heavy rain...
Of the whole spectrum of light, why limit your self to the visible spectrum
Cost and complexity. Tesla used to use radar as well, but went camera's only as sometimes the radar and camera's conflicted. Ultimately it was the camera's that were being proved correct so Tesla decided to go the camera only approach. I must say seeing what Tesla are achieving with V12 of FSD, I really think they may have cracked it.
Imagine a world with perfect self driving tech, no one should buy a car, let the robots drive
Dude that's really boring,don't you want to drive sometime?
Imagine a world with perfect public transportation
@@carson9903 If by public transportation you mean individual cars with my own space
Of course we need self-driving cars….there are too many distracted drivers on the road these days…..every single one of them is looking at their cell phones.
Things not addressed that are interesting aside from sensors:
- Usage of HD maps
- Comparison equally to number of miles driven on self driving auto pilot vs cruise control auto pilot.
- Volvo including single LiDAR sensor on new electric cars for partial self driving on highways.
- blue cruise and super cruise.
Very shallow comparison I would say. There are a lot technicalities when choosing sensors. Comparing accidents is the dumbest thing I have seen. If waymo had solved self driving through LiDAR, $6000 would be pennies for the gains that we could get. However, it's not that simple. I'm not saying LiDAR is not a good sensor. But I don't think creating realistic map of surrounding though point cloud is as necessary as people think. We can make a good enough 3D image of the surround by using multiple cameras. Software is the main bottleneck right now, not the sensors. If software reach at the level where hardware becomes the bottleneck, then we can talk about LiDAR.
Great series!
i just looked up on google, "75000 to 6000 a pop" actually meant 6000 per day. They way he cover this video is a bit bias when talking about lidar and tesla. he covers all the details for lidar and pros, but never covers tesla's.
It reconstruct the debate of bicycle that we need more than 2 wheels to avoid falling.
Is there a reason you chose not to compare the number of crashes per one million miles for Tesla like you did with Waymo?
Lidar doesn't even work in heavy rain. So all companies that trust on lidar must have a 100 % working redundant camera system if they want their cars to drive autonomous in all weather conditions. Why should anyone develop two separate systems if only one of them is capable of mastering all situations? For me Lidar is only a historical intermediate step, due to the lack of skill in processing 2d camera data to 3d data. Humans only have a visual sense as well. And humans never make accidents because we lack a lidar sense. We make accidents because we overestimate ourself (driving-skills, alcohol, drugs etc.) or we get distracted. Both will not happen to computers. Since FSD Beta 11.4 it's not about what sensors anymore, but about how much data the AI has to be fed to drive better and safer than human.
Thumbs down. You don't give your stats in accidents/miles driven. You can't compare the numbers of accidents without a a common denominator. This comparison is misleading and that makes me think The Verge was paid by WayMo to mislead the public. Typical.
As an editor, I know the one on this video had fun with those two shots hahaha 6:50
100% self-driving cars were a fascinating dream that will come true "next year", maybe 5-8 years ago... Now, some 8 years later, it's like that fancy uni degree you fought SOO hard to get! Received 50% of it. Getting 50% of the desired salary. And paying the student load for the next 20 years...
Yeah Elon been saying for 10 + years now that it's ''one year until we have FSD''. It will take decades before we can really trust it. Especially in Europe
I thought Elon was pretty clear that LIDAR / radar was also being removed because it was too low resolution and interfered too much with what the cameras could see and caused conflicts with decision making. Now that higher resolution radar tech is becoming available I thought they were adding that back in.
The problem with decision making is just one reason most companies will toss Lidar. The main reason is that lidar won't work in all weather condition. So they have to set up a vision system as redundancy. And as soon as they have their 100 % vision system working, they cut Lidar. Pretty logical.
Exactly low resolution radar can cause a conflict between the camera input and the radar input. Which input do you trust? ... So to prevent phantom breaking in the past it would just do nothing.
So, no consensus what tech is best for FSD, still we expecting LVL 5 anytime soon
The whole driver is in charge, can be applied to cars with Lidar also. No matter how cutting edge unless 100% safety can be garanteed drivers are always in charge unless it's like a driverless taxi, which are extra precautious. They stop even when not needed just because it's trained to be so.
Lidar absolutely helps and is better than Tesla's solution. Tesla's camera based solution has many faults which Lidar solves. On top of it, they processing isn't as heavy as you would think.
The data from the Lidar is fed into a ML model. These models although heavily resource and time intensive to train not so much when being run. And more over there will be chips that transform the Lidar to a more easily accessible and usable data. Apple is already doing it with the R1 in Vision Pro. These cars all have it.
I would trust a car with Lidar more than one without it. The depth information it provides cannot be matched with a system reliant on 2D cameras. The best it can do is add another sonar sensor to eid. But since light travels faster, that also wouldn't be a direct replacement for Lidar.
If Lidar is so amazing why is Waymo not national yet after decades of development? If you are going to make an honest comparison of the two sensor choices you cannot just mension cost.
Lidar cannot distinguish between some objects or read signs and it doesn't work well in certain weather conditions just to mention a few. There is a lot that goes into the consideration besides the cost.
That's why Waymo doesn't have just Lidar, they have a full set of sensors including cameras. Elon simps are so freaking stupid 😂😂😂
Lidar can only get range data, nothing more. You still need a 2d camera to provide the system with enough data to work with.
Bait and switch....Autopilot accidents....but you didn't mention statistics from FULL SELF DRIVING.
I think LiDARr is important but only for additional context, the full package of 360 is basically overkill...
but Tesla's approach allows the system to react to the "event" than to react by the "event" ... liDAR is basically just checksum with this approach...
i dont understand whats wrong on "laser is the sause"
And of course you did not mention occlusion of LiDAR in harsh environments. It becomes useless in rain, snow and fog.
Neither tech is a "magic bullet". They're complimentary, not redundant.
does cameras excel in those conditions?
@@daviidoncompared to LIDAR. Yes. But neither is “good” in the rain
@@daviidon Camera with radar (existing technology) is far more useful than LiDAR in these conditions.
Arguing one tech over the other is responsible. It likely to be mix of the two.
Also, there is a thing called "solid state lidar", much cheaper.
3D Lidar = capture depth info
2D Cameras = interpretations i.e. require AI guessing
Lidar, radar, and cameras. All are needed to have safe fsd, it wont be cheap though
I was hoping to actually hear from experts. We know we were overzealous think lvl 5 would already be here, but we still cant decide what tech is required to accomplish this?
id say yes and no to the title question. hes right to train without lidar, but once its good like that adding lidar would be perfect.
Great video!
the tests works because it's phenix, a very simple grid, and not alot of drivers
Who think safety was “expensive” was the guy in Titan
The moment it rains or so Lidar is rip so yer. There you have it, very easy to say witch one will win.
Having more sensors without adequate “brain” tech. doesn’t necessarily get us to true autonomy/robotaxi; in addition to more Teslas being on the road (which you rightly pointed out) Tesla FSD Beta (emphasis on BETA) also drives on highways with higher speeds, thus if crash occurs, more serious damage
Telsa cars with Ouster Lidar was spotted on street
4:45 LIDAR now $6K, down from $75K
Quantum level AI should be able to use cameras and infrared cameras...Lidar might be better but optimizing AI should be similar...It's like with the Canadian fires, that stuff is arson and needs quicker response....AI cameras in balloon drone hybrids with infrared cams...both technologies need to advance, so even if Tesla is wrong, cameras with AI plus infrared tech need investment advancement anyway.
I miss the Tech Verge guy
We all know Elon removed LiDAR because it's expensive. With that said, however, comparing the number of crashes something like Waymo has to Tesla's cars is like saying a starting pg who plays 38 minutes a game and has 4 turnovers is more reckless with the ball than a center who plays 7 minutes a game and is only put in to play defense. Not only is the Waymo geo-fenced, it's also mapped. It can travel routes that have been mapped before. For the most part, it also can't travel on highways and is speed reduced. It's not put in nearly the same situations the Tesla is.
Again, I really wish Elon would pu the LiDAR sensors back, along with the ultrasonic sensors. I just don't like the crash number comparison.
There was never any LiDAR in Teslas. Just radar that was removed earlier this year I think.
@@MockarutanRadar coming back with HD radar
Came here to say this, so thanks. The comparison crash data was misleading at best.
Not mentioning miles for tesla self deriving was verry misleading
There can never be fair comparison between the two unless Tesla is given the same opportunity to do full auto drive.
Having a human driver, who understands how good Tesla drives itself in certain situations, decides when it's best to do auto drive and when to intervene is not the same level of full auto drive on the street no matter the situation.
The milage difference can never be truely compared when they are doing different things in those milage.
n what happens if the internet goes down ..... ?
Forget self driving cars - give us self driving buses and autonomous subways / light rail. That’s how we could actually solve many of our challenges-climate change, urban sprawl, public health problems, social isolation/disconnection, etc.
“The company reported “ 5:39 “federal crash data” 6:00
Please provide better data citation
More to note, he mentioned auto pilot, which is the one used on highways, while compared to Waymo which only works on specific city roads, that alone makes this comparison silly.
All the sensors on waymo isn’t what makes waymos drive better than a Tesla in the 2 cities waymo works in. It’s the HD maps that tell the car every move they have to make. Where as Tesla doesn’t want to really on expensive HD maps that take an enormous amount of time and energy to maintain. Tesla has basically already solved the computer vision aspect eliminating the need for all the sensors that cost hundreds of thousands per car. 6:38, you compared Tesla autopilot and waymo, autopilot and FSD are wildly different things. FSD is way more capable and can see everything, autopilot cannot
Lol, reporters should not have polarising opinions
Forget about self driving. Radar has been reliable for basic collision avoidance and even Toyota Corolla now have it. Tesla removed it and now the cars are plagued with phantom braking and not failed to brake in actual obstacle or human. That is possibly a major factor for higher insurance cost to Tesla owners.
im confused you just said that full self driving cars work with cruise and waymo and half the vids you showed of them not working were of cruise and waymo in San Francisco lmfao
I’d rather have better public transit than self-driving taxis.
I’m going to LiDAR in macaroni from now on.
Did way more partner up with Jaguar?
Five years back GAN & Generative AI (AKA Text to Image AI) are far from deployment. Now our mobiles are capable of running this AI and AI itself has ground breaking applications.
Cameras are also upgrading at rapid pace, exclusively stereo cameras, night vision camera tech, sensor fusion, HD Maps are maturing at very wisely. In future, LiDAR might be just an expensive sensor with extremely large amount of data to be processed per frame. Computationally processing LiDAR data might consume huge power and also impact vehicle's range. LiDAR has huge applications, but stereo or bi-focal cameras will be able to produce similar depth information in near future. They are cost effective solution and when was the last time you got into a vehicle without a camera?
Thats not how it works. Lidar consumes much much less computing, because it does not need to figure out how far away something is, it knows it immediatley. The returning laser beam has the exact distance measured.
Cameras has to use multiple pictures, over a time, using stereoscopic vision to try to calculate fistly what objects are foreground, and what objects are backgound. And then try to figure out how far the background is, and what direction objects are moving. This consumes tremendeous amounts of data. Lidar gets all the answers in an instant. Super cheap in data compared to cameras.
@@intelligentunite4557 Do you thing LiDAR based object detection is as simple as in cameras?
@@kishorekurapati9172 No one is saying to replace cameras with lidars. With lidars you can easily tell if an object is big or small, far or close. You can then pass on this information to cameras. It's much safer for a car to know that an object is far rather than "think" it's far.
now they renoved USS
I don’t see the point with self driving cars over shorter distances through cities for example. Long distances on a highway, yes. I think it’s a solution to a problem that doesn’t really exist. Some use cases wound be good though, for example allowing the disabled to get around.
They intentionally do not say how many miles Teslas have driven. This 1 million miles for cruise is laughable, since Tesla has more incidents but has trillions of miles driven.
Trillions??? In full self driving mode???? I think not.
Cameras are higher def than lidar.
Get back to me when these lidar cars drive faster than 35 miles an hour.
It should be miles per crash/fatality.....not how many.
Wow this video is on the 'verge' of being a hit piece. No mention of the Tesla occupancy network as a replacement for LIDAR? Comparing accidents from 1 million miles of driverless on curated pre mapped roads, with many million human controlled autopilot miles on all roads, without mentioning any FSD miles? Very weird.
When did Tesla remove LIDAR from its cars?
Tesla has never used Lidar.
Compare properly many crashes on teslas side but you fail to mention that they have millions more miles driven
Just realized this video is pre-supervised-FSD. FSD 12.x (AI version) kick lidar's behind.
As a computer scientist, I believe that more data is better than less data. The data provided by cameras and the data provided by LiDAR is different and neither product gives the “full” picture, making these two technologies complementary instead of redundant.
Initially, Elon was against LiDAR due to the cost, size, and aesthetics of the equipment. During the past decade, the technology has shrunk in size and cost so to the point where they live on our iPhones, so I’m not sure why Elon still opposes the technology in 2023.
Perhaps if Elon spent less time jerking himself off on Twitter and more time reevaluating technology trends and waking up to the fact that Waymo is leaving them in the dust, then maybe Tesla can experience some forward movement again and keep the promise of FSD that Elon has been pedaling as only 18 months away for the past 10 years.
The fundamental question remains, does it work in SNOWY/MUDDY conditions! No!, radar tech will always prone as all other visual optics solutions will only be used as complementary devices no matter what people say. Yes there will be SNOW or MUD covering those pretty sensors on your cars no matter where you live.
Well your video are comparing Waymo and Tesla, right as you mentioned. Why didn't you show 1 million miles driven by Waymo vs Tesla FSD and jump on a 2019 comparison. Also did FSD get worst now compared to 2019? Is that a really a good assessment? Not really.
Dissappointed in the comparison of crashes between tesla and lidar vehicles. This must be normalized for miles driven to be a fair assessment.
I’m not so sure, with added cameras to Tesla’s HW4 I think they could get pretty darn close
I'm so relieved that Lidar vehicles shift liability to the auto manufacturer when it runs over a kid at a crosswalk. Venture capital spending is billions of dollars to put cab drivers out of work. The government is investing hundreds of millions of dollars into autonomous vehicle research instead of on quality public transportation. Our priorities should be people, not tech.
Anyone paying any attention already knew this about LiDAR and Self-driving… was so counter productive when Elon turfed it, but the costs explained it. But it’s good to have it all wrapped up so nicely in one video.
Elon said Lidar was not needed. Seeing V12, I think he may well be proved correct.