Nice summary of LIDAR technologies -- is it a "crutch?"

Topic: 
Lidar report details all the players

Earlier this year this nice summary of LIDAR companies was published. While it misses a few projects, I recommend it as a nice visual overview of the many LIDAR technologies in production and under development.

This should make a stronger case on the question of the role of LIDAR. Almost all teams are using and planning to use LIDAR, expecting some of the companies in this report to deliver production quality, reasonably priced LIDAR around 2020-2021. A number of teams, namely several startups and Tesla, are trying to work without LIDAR. As I have written before, this is an error, because it is usually done with the goal of saving money -- cameras are much cheaper, especially when you want to look in all directions. Now is not the time to be cheap.

Elon Musk makes one of the few counter-arguments, claiming that "LIDAR is a crutch." That it is just letting teams get away with not making their vision systems sufficiently good. He feels it is making teams chase a "local maximum" though he may have mostly been critical of approaches that rely almost entirely on LIDAR with minimal use of vision. That is not the approach of any major team today I am aware of.

Tesla's lack of LIDAR has a more mundane explanation -- there is no suitable LIDAR on the market today that you can buy in production volumes for the cars he is making. Tesla is young but already has a legacy -- it is making and delivery cars today, and has prime focus on what it can do in the cars it makes today. Those cars can't have LIDAR, so their team works hard on what they can do with what they have.

For a while Tesla would let you buy "full self drive" as an option when you bought a car. You paid for some extra cameras and the promise that later, a software update would give you full-self drive. That option was removed last month, indicating Tesla has realized that the existing sensor package is probably not going to do it and it was silly to sell a function when you didn't know how you would deliver it, or if you could.

Today, every team has just one focus -- get safe enough, and prove you have gotten safe enough, so that you can get out on the roads developing a real service. They all want to get there soonest, and if they are smart, they are not trying to save money doing so if it costs them time. If it costs you capability, it costs you safety and it costs you time. Saving money is for 2025, not today.

LIDAR's superhuman sensing is no crutch. While far from perfect, it is never going to have a problem telling something close from something far. It is never going to have a problem seeing at night. That's too much to throw away. Fused with vision and radar it's a win. The main competitor to it would be high resolution RADAR which would do almost all LIDAR does but also see through fog. (Radar will probably not see road texture and line markers, however.)

Comments

Tesla still lets you buy FSD, they just took it off the menu to "prevent confusion". It's unclear why this confusion took 2+ years to surface.

Will Quanergy ever deliver, or are they mostly hype? Seems like the MEMs guys and cheap spinners like Ouster will take the Robotaxi market, since they can actually deliver something.

I no longer get inside information on Quanergy. They have been ramping up production, which nobody else has done, but they aren't selling a large volume suitable sensor yet. They did just do another funding round at $2B valuation.

So is anybody buying full self drive with it off the menu? I recently ordered a Tesla. I would not have ordered FSD if it were on the menu both because I don't believe they are close, and also because it's not that much more to buy it after if they do get around to delivering. My belief is that if and when they get it, it will need extra hardware, even for people who bought it before, so the main benefit those people will get is that Tesla eats the cost of that new hardware. Perhaps that's a good deal. If not that many people bought FSD, then Tesla might well decide to just take a large loss on the small number.

Also, FSD is software only. No cameras are added.

I would have sworn that Tesla said if you bought FSD they installed extra cameras on your vehicle, and promised you future software.

I think last time we went through this you pointed to Tesla's statement of "This doubles the number of active cameras from four to eight."

Key word there is "active." The hardware is the same, but the extra cameras aren't active if you don't pay for the upgrade.

"All Tesla vehicles produced in our factory, including Model 3, have the hardware needed for full self-driving capability at a safety level substantially greater than that of a human driver." https://www.tesla.com/autopilot

However, they also have said they will be adding the new processor to cars that bought full self drive.

Nothing about installing more cameras though.

Though we agreed up thread that this was an older bit of information and now they claim the cameras are all there.

My prediction however is that it is likely that when Tesla first offers to turn on full self-driving, it will involve a hardware upgrade package (including a lidar or high resolution radar) which they install in your old car, at a loss, for those who paid in advance for FSD. And that they will break the promise about what it costs to "add FSD later" with a possible lawsuit over that. If the lawsuit wins, they may have to withdraw FSD from the market. I don't know the legal history of doing something like they have done when the car order form said, "You can add FSD to this car later for $6,000" or whatever the price was. What if they can't do that?

(Fortunately for them I don't think it's too likely, as a full self drive hardware package eventually comes down to just a few thousand dollars, so they can put it in with not much profit on the old cars.)

I wonder if it's really needed a 360 degrees lidar for an autonomous car. I could be more cost effective to have a front 60 degrees front lidar and use cameras (plus short distance radar?) for other directions.

You need to be fast and very safe in detecting front features, but lateral and rear features requires a less performing system.

Because LIDAR is still costly, what you describe is a common plan. If LIDAR gets cheap, you might as well have it or something like it in the other directions. When I say something like it, there are short range 3D sensors which cost much less and some people make use of. You don't need super long range in every direction. Though you do need to see cars coming up fast behind you, and cross traffic.

Add new comment