Earlier this month, as Tesla's robotaxi fleet in Austin extended its hours to 11pm, Elon Musk talked about his fear of the night: "Literally trying to avoid grey kittens on grey tarmac in the dark." That's the use case for lidar, the laser sensor that most autonomous-driving companies use. Musk called it "a fool's errand" in 2019, when a ubiquitous Velodyne lidar cost around $75,000, and doubled down on cameras alone. So far, the bet looks good. Tesla's FSD is the best of its kind, and by many measures ahead of Waymo's. It has sold 1.48 million FSD subscriptions, and when a Chinese car media company pitted 36 cars against highway obstacles, Tesla's camera-only Model 3 and X achieved the best score, beating several lidars. Even Waymo co-CEO Dmitri Dolgov agrees that "Humans of course can drive with just eyes." But lidars have come down in price dramatically. BYD, China's biggest automaker, includes lidar as a $1,700 upgrade on all its cars, even its budget Seagull, and its chairman Wang Chuanfu says that's "already at cost". Pony AI, a Chinese robotaxi company, says its total robotaxi cost, including lidars, is around $33,000, "cheaper than the lowest-spec Tesla Model 3". Not every Chinese automaker is a fan; Xpeng chairman He Xiaopeng has dropped lidars on his newer cars, saying "in cars it is no longer needed". But Chinese rules for the most advanced autonomous cars require more than one kind of sensor, and as of this spring, Baidu's lidar-equipped robotaxis had driven around 137 million driverless miles. The cat problem is a rebuttal, too. In October 2025, a Waymo with lasers hit and killed KitKat, a gray tabby cat who lived at a corner store in San Francisco's Mission District. And Musk told Electrek privately in 2021 that "vision with high-res radar would be better than pure vision." Cameras alone are good enough for most driving. And now that lidars may cost $200, the question is whether they're unnecessary ballast or cheap insurance for that once-in-a-while emergency on a dark night when you really need it (if vision alone can't figure it out). Images (cropped and graded), via Wikimedia Commons: Waymo Jaguar I-Pace — 9yz, CC BY 4.0; Apollo Go RT6 — S5A-0043, CC BY 4.0; Velodyne lidar — Steve Jurvetson, CC BY 2.0; Tesla cameras — Votpuske, CC BY 4.0; Dmitri Dolgov — Collision Conf, CC BY 2.0; Wang Chuanfu — Alexander-93, CC BY-SA 4.0.
ADVERTISEMENT
The cat .. always the cat . From quantum physics to driverless cars .
You didn't mention that the cat started sleeping (not actually sleeping) under the Waymo and no average driver would have been able to know a cat quietly slipped under his car
From working in the field, LIDAR does more than just find grey cats on grey pavement. Computer vision is scale ambiguous, affected by weather visibility, and still requires massive data streams. LIDAR solves the problems with OCR, so they are complementary systems working best together.
Problem is not the cost of lidar! It's the massive data it generate, it's mostly useless on cars without a 10-20k computer
The irony that the one method that would be more suited to avoiding the grey cat incident is the one that affirmed the danger
I also don’t understand why not just do road-bound infrared like. Waveguided infrared light
Part of the issue with Tesla crash statistics is that they turn off self driving 1.5 seconds before impact so that it is not engaged when the accident occurs. They also don’t slow down for flashing emergency lights. Also if a vehicle has turned on it’s side it does not get recognized, so it will hit it at highway speeds without ever braking. The one that repeatedly kills people is when there is a divided lane on the left (ie hov) with a yellow chevron symbol. To see an example look at California southbound 101 hov lane where 85 splits off near mountain view. I suspect that Tesla sees it as two up arrows and goes forward to impact it. For this specific intersection, the state of California sued Tesla. I don’t follow Tesla accidents, but I have read about 2 fatal Tesla accidents with the chevron symbol dividing an hov lane in the last year.
I work in robotics and deal with all of these issues. It’ll take a unified multi sensor suite to solve these issues. Humans use multiple unified sensors, robots navigate the same 4D space so….
Also you clearly have no idea what you talking about the comparison with lidars
RIP Kit Kat ❤ We will remember you
Who on earth said Tesla FSD was better than Waymo’s?
Don't get me wrong when I say this. I'm definitely impressed with Chinese auto makers right now. But unless things have changed, last I read they were all heavily subsidized by the government and none are truly making a profit. Which is why they can offer all this tech at or below cost. No clue if there's a price difference before tariffs/taxes when sold in other countries. Unfortunately I'm stuck driving pick ups so I haven't spent any time looking at the pricing.
Issue with Lidar isn't about the cost but the possible conflicting info being sent to computer on board and having to figure out which input to prioritize
I think Musk uses LiDar in Space X missions for docking to ISS so he clearly values the technology.
Infra Red/IR is very Blurry when weather is Fogged or heavy Rain. dunno why people suggest it. beside Radio Waves system also has advantage that it can measure Range
By 2030, every self-driving car will have enough cameras and lidars to be able to navigate even the darkest corner of the world. Lidar + cameras are the way to go in the future of self-driving cars.
No. Humans can't drive car with just eyes. Ask anyone who has driven in whiteout, thick fog, heavy rain, against sun or any other visual obstruction and they can tell it's hard and even stupid. We just trust that we can do it.
Get a heat sensor while you are at it
Both. Both is good.
I understand why not at first but being stubborn keeps us in the past.