There is a lot of love for LIDAR around here. What is missing from the discussion is whether LIDAR is worth it. Early on Tesla used LIDAR and rejected it saying that the added complexity was not worth it. I accepted the argument thinking they know more about it than I do. I no longer paid much attention to LIDAR.
The recent posts suggest that since so many others are using LIDAR, Tesla must be missing out on a good technology. Tesla does use LIDAR for research but not for production, Why not for production, there must be some explanation. Yes, money!
Elon/Tesla are penny pinchers. When you save a penny on a million Teslas that’s $10,000 in free cashflow. When you save $100 on LIDAR that’s $100,000,000 in free cashflow and net income.
How much does LIDAR cost?
Here is a breakdown of LiDAR costs by application:
Automotive and robotics
Entry-level autonomous vehicles: Simplified, less expensive sensors can cost between $150 to $300 per unit.
Mid-range autonomous vehicles (Level 2+ to L3): Expect to pay $600 to $750 per sensor, with the total system cost often exceeding $1,500 for multiple sensors.
Robo-taxis and high-end autonomous vehicles (Level 4): These sensors feature high reliability and a 360-degree field of view, costing between $1,500 and $6,000 per unit.
Google AI
At $1,500 per million Teslas that’s $1.5 trillion (with a “T”).
No one is disputing that LIDAR is pricey. You understand that, correct? We are disputing if LIDAR is necessary for full self driving. Tesla has yet to actually pull that off. Waymo has.
Related to this, if LIDAR is in fact necessary, but is also too expensive, it means the dream of cheap robot taxis is just that, a dream. A big, wasteful dream.
Elon still doesn’t have taxis without safety drivers, and has yet to show that they don’t need them either. Elon also said 50% of the country would be covered by his service by the end of the year. I’ll wait…
Lidar does not work. Tesla ran a bunch of tests between vision and Lidar and radar. They found out that whenever there was a conflicting signal, vision was more accurate and reliable. Moreover, vision is also getting progressively better much faster.
All automakers are penny pinchers for the reasons you mentioned. The parts cost of even complex components like airbags is surprisingly low.
That said, a common occurrence in these threads is that Tesla proponents compare Tesla’s promised future tech with its competitors’ current tech, and surprise! Tesla always wins.
A much more useful comparison would be with both Tesla and its competitors’ existing tech. A while back I heard and interview with Waymo co-CEO Dmitri Dolgov who said that Waymo Driver works with vision only, but it doesn’t meet safety standards. From that we can conclude that Waymo is studying vision only and may be headed in that direction ultimately.
Another company whose vision-only system doesn’t meet safety standards is Tesla. Tesla has not introduced an L4 system anywhere, and as far as I know as not even entered the regulatory process to do so. Waymo on the other hand will be in about 10 cities by this time next year, one of them international. Robotaxi has been central to Tesla’s business strategy for a decade now. If they could be doing it, they would be doing it.
So if we compare apples-to-apples, Waymo’s technology is more advanced, but more expensive than Tesla’s.
There are a couple ways this could play out. One is as the Captain suggests, Tesla figures out the L4 problem and now is producing capable vehicles in large volume at low cost. But another scenario is that Waymo figures out the L4 problem first (on the consumer level), licenses it to other manufacturers who can then offer it as a option that Tesla cannot offer at any price.
Or both Tesla and Waymo solve vision-only at the same time. Or by the time Tesla cracks vision-only, Waymo already dominates the US robotaxi market and Tesla can’t overcome the first mover advantage. It is easy to come up with lots of other scenarios too. The least likely scenario is that the cost of a LiDAR suite remains frozen in time.
Tesla makes their own cars. They have their own inference chip and related hardware onboard. They have OTA infra. They have gigafactories in multiple continents. They are on a path of making cybercabs which will have no steering.
Tesla makes 35k cars every week all of which has the FSD hardware inbuilt. Waymo has total of 2000 cars which have been fitted with Waymo hardware with ongoing expansions and manufacturing to support growth toward 3,500 by late 2026.
…but none of which are yet capable of driving themselves without a Tesla employee in the car, save for the occasional one-off delivery (which doesn’t seem to be repeated).
Which was the point. Your elucidation of Tesla’s advantages is entirely premised on Tesla’s existing hardware someday being capable of actual self-driving. And perhaps one day it might. But that day is not yet come. It depends entirely on a software improvement that may or may not happen. We might be months away from having software that can use the current FSD hardware to self-drive - but that day might instead be years away, or never (as with all the HW3 cars that will never self-drive with the hardware they were installed with)..
Comparing what the cars are currently capable of, Waymo’s vehicles are capable of regularly being operated autonomously without an employee in the car, and Tesla’s are not.
A key difference is that no a single one of the 35K cars per week can legally drive themselves. That’s a key stumbling block if robotaxi is central to your business plan.
Neither of those statements is accurate. “Each car” does not have a remote driver - these cars are not teleoperated, but rather have access to humans in the event that the car is unable to proceed. And even then, it’s not a real time intervention.
And the car doesn’t have $200K worth of hardware on it. Some time ago, the cost of a Waymo car - both the car and the hardware - was estimated to be around $200K. But again, that includes the cost of the car, not just the hardware, and the costs have certainly fallen since then, as LIDAR generally has gotten cheaper.
I mean, it’s working better than Tesla’s project - which does not yet function at all without a human employee in the car. And it’s scaling pretty well, so far - they have far more vehicles that can actually drive themselves without a human employee in the car than Tesla does, and that number is increasing at a substantial pace.
Of course I understand that LIDAR is pricy and Tesla does not think it’s worth the money. That was the whole point of the OP.
Three fewer zeros? LIDAR made me do it!
Having read all the posts the consensus of the opposition is that if Waymo is doing LIDAR that’s the right way. Tesla has decided some years ago that LIDAR just adds complexity with no real benefits. If the benefits aren’t there then LIDAR is redundant. and a waste of money.
Tesla’s FSD has been generating revenue for many years now and cumulative revenue is ~5B (and compay’s TTM revenue is ~100B). Investors see a path to autonomy and market cap has reach $1.5 Trillion.
Waymo’s cumulative revenue is ~500 million and it has spent roughly $30 Billion so far from big daddy Alphabet / Google.
Tesla was formed in 2003, IPO’ed 15 years ago. Waymo was formed in 2009.
Yes, but the comparison wasn’t whether “investors see a path” to some future technology that doesn’t exist yet. The author was talking about what tech the two companies have today. Waymo has technology that is expensive and limited to a few thousand cars, but actually works to provide autonomous rides to people without having to have a company employee in the car. Tesla does not yet have technology deployed to do that.
They hope one day to do that, but right now they don’t have a software package that actually can do that. They project that the hardware in the car is strong enough today to do that if it gets the right software upgrade…but they were wrong about that with HW3, despite their insistence to the contrary, which certainly means it’s possible that they’re misjudging whether any of today’s cars can ever do that.
Sure. And part of what helps you assess the likelihood of those future cash flows is what the company currently has been able to do. Tesla has not yet been able to develop the product they are counting on to generate all those future cash flows, despite their confidence that such product would be available within the next year for the last six-seven^^ years or so….
^^Sorry. I have a 12 year old in the house. Everything is six-seven…
That is not correct. The consensus is that the claimed benefits of Tesla’s camera only approach have not been proven. For example, you claimed that Tesla’s approach would lead to dramatically lower costs. But since Tesla’s does not have a L4 system we can’t know if that is true or not.
Tesla’s robotaxi does not operate in geofenced areas because it doesn’t operate in any areas.
It doesn’t operate only in good weather because it doesn’t operate in any weather.
Tesla’s robotaxi’s top speed limit is zero mph.
Tesla’s robotaxi doesn’t go on freeways. It doesn’t go on surface streets, it doesn’t go in parking lots, and it doesn’t operate in Boring Company tunnels.
Other than not actually existing, it is superior tech.
