Tesla vs LIDAR: The Battle of Self-Driving Cars

Tesla's CEO Elon Musk hates LIDARs and believes that using LIDAR tech for self-driving cars is "freaking stupid" but pretty much all car companies working on self-driving technology convinced that LIDAR is the answer to achieving Level 5 autonomy. So, who's right?

OUR SPONSORS:

IMPORTANT NOTICE: Sponsors of this channel do not in any way control or influence the opinion or views expressed in E for Electric videos. All sponsor content is clearly marked and announced in every instance. Views and opinions expressed on this channel do not necessarily represent or reflect the opinions of our sponsors, their partners or their associates.

FAQ: How can I be impartial in covering EV brands if this channel is sponsored by one of them?

PATREON PAGE: Watch "E for Electric" LIVE and get more exclusive content while supporting my my channel at:

Electric Car Apparel & Merchandise from E for Electric:

Our website:

[GreylanderTV]

If it is unnecessary then even if price is down to $500/car, then that is "pretty stupid", as that is a huge chunk of profit or a huge chunk of savings for the consumer. LIDAR gives you a high-resolution depth map plus doppler speed. Computer vision(video) can provide the same information in a different way. Ten years ago, depth perception via computer vision took too much computing power, so LIDAR was the "easy" way. It did not help with driving much, because you also need the same (or higher) levels of computing power to process those depth maps to recognize objects and make decisions about how to drive. Today we have those levels of computing power, so LIDAR is redundant. It may do some things _slightly_ better than computer vision right now, but that is changing fast.

Furthermore, when it comes to training AI, it is far easier to "label" images, which humans already intuitively understand, than to label the 3D depth maps provided by LIDAR.

Another way to think about it -- there is no way to have level 5 self-driving without fully solving the underlying AI problem of computer vision, and once computer vision is fully solved, LIDAR is simply irrelevant.

Investing in LIDAR would just have made Tesla's cars cost more and been a hole into which they dumped R&D money.

[CSHarvey]

"Does Self Driving require LIDAR?"
Wrong question. Does Lidar-based self driving also need computer vision? The answer is yes. Lidar can't read visual information like road signs.
But if you've solved for computer vision to process VISUAL information.. then you've also solved for range and distancing, at which point Lidar becomes redundant.
This answers the reverse question: Does vision-based self driving also require lidar? And the answer is no.
Can you do L4 without computer vision? Sure. Can you do L5 without computer vision? Possibly.. but at that point you'll have already needed computer vision so we're back to Lidar being redundant.
If you're building purpose built robotaxis that will be geofenced to hypermapped urban areas where all navigational signs are mapped and observed, you can operate using L4 Lidar self-driving. But it can't function without a significant amount of overhead maintaining all the navigational metadata.
Conversely, a vision based L5 system can operate without a constantly updated navigational metadata system, but performs even better with it.
So, where your argument is 'Wouldn't it be good to have Lidar as a backup?" It would be better to have the navigational metadata that the lidar systems rely on to function as the backup than to have the lidar sensors themselves.. And again, you can build and maintain that system from computer vision.. making the lidar sensors themselves double redundant.

[dwaynecranston]

With your tea, would you like a chocolate croissant for only an extra $10, 000

[garychan]

when elon says, "vision", he's not talking about
He's talking about the AI understanding of visual....

[TheScottShepard]

Because rain, snow, and even fog blind lidar. That’s why its ok to use it for docking to ISS. Very little weather up there

[robinholmes]

I get the "why not" argument, but I think you are looking at the problem backwards. The Lidar graphics you used were not real, the actual info is much messier. Lidar, like Radar, gives you activate point and distance info but very little clarity so it is impossible to be sure what you are seeing. It is easy to get, with GPS and Lidar, a functioning control system and a basic level of control in ideal circumstances, but our roads are often way off ideal! You reach a technology brick wall, you need to start again with something that can give you the required clarity, cameras!
With a great deal of hard work, data, computing power and self-learning, camera-based control can not only correctly work out what objects are and where, but also forecast in what direction and how fast they are likely to move. This is the ONLY way you are going to be able to have a self-driving system that can avoid, say a cyclist on a curve, at night, (Wamo!).

[darbyelliott]

If Tesla doesn't need LIDAR and they make FSD work without it.... then who cares if they use LIDAR???? Just get FSD working, imo....

[marcuschristian]

Despite all the economic crisis this is the right time to start up an investment

[NotFinancialAdvice]

Well... I'm watching Elon's FSD beta drive in traffic basically anywhere in the country. And the LIDAR stuff... is well... you don't even see those videos.

[talkingBS]

if lidar is shooting infrared laser light and receiving the signal to measure... i wonder how it operated if 10 lidar cars are on same roadway... can it differentiate its own lasers from other ones? curious...

[jamesbuchanan]

Human's have two eyes and no LIDAR, and diligent human drivers are capable of doing an excellent job.

(Cameras are multi-sensors--in a massive sense--with each and every pixel element being an independent, multicolor sensor. When multiple cameras are combined into stereo vision, it is a terrific way to go!)

[_percent]

I am more excited about Tesla's future than my own

[kimlynch]

Hi Alex, I generally keep quiet but on this I am speaking as a design engineer and programmer in military aerospace, with very little knowledge in self driving. I have listened to Andrej Karpathy and done some light reading on subject. I will say while I believe with what is out there they are going to achieve level four and later level 5. I can't be 100 percent sure. For us to answer this we would need to actually be involved in what they are doing. Karpathy talks about how they achieve the 3D and we have seen what is captured by the sensors. I think the Tesla seen running a lidar rig is probably helping verify and calibrate the system they have in cars. Also remember we humans don't have lidar built in and do a pretty good job. I think the multiple sensor suite is probably just as capable.

[skirkwood]

It's not just the price of the lidar unit, but also the fact it uses 10 Watts of power. You're gonna need a bigger battery, or have less range. Waymo has 3 lidar systems installed.

[Dadinald]

You’ve become much better at puns! I like that! 👍🏾

[barton]

Yup. No such thing as too much information to make self driving vehicles safe.

[GH-uqwr]

Bro we are not socially distanced, you're 12 inches away from my face.

[radudragan]

I see...so less redundancy is better than more redundancy?
Call me a hoarder, but when it comes to safety, I'd rather have it and not need it, rather than need it and not have it?

[ws]

The real objections to Lidar are cost, universality and "where do we put it". Cost is obvious. Until recently, automobile capable Lidar (100 foot range, minimum) started at $25, 000. By universality, I mean Tesla's desire to have the self-driving hardware setup on every car. And as for "where do we put it", have you seen the ugly package on top of Google's self driving cars?

[LewdCustomer]

Is "friggin' stupid" some kind of code. Elon should just say what he means.