Hacker News Reader

Comment by redbrickroad | Hacker News Reader

At the current cost of a self-driving car (~$250K), it would be ridiculously expensive to build a fully self-driving ride-sharing fleet [1]. So expensive that I doubt it's possible to make real money that way right now.

Self-driving cars are a long game until companies can bring down the costs. It might be years before that happens. Uber has the advantage of being able to ramp up self-driving cars as part of their existing fleet until then.

[1] https://qz.com/924212/what-it-really-costs-to-turn-a-car-int.... [2] https://www.uber.com/newsroom/company-info/

Edit: math mistake

View on HN

In the article it says WayNo has already reduced the cost of Lidar by 90%:

“Waymo, the Google spin-off, claims it has cut the cost of an experimental version of the high-end LIDAR to around $7,500. It did not release details”

More details here: https://arstechnica.com/cars/2017/01/googles-waymo-invests-i...

They "reduced" the cost by stealing Velodyne's Lidar technology and filing a false patent on it.

https://arstechnica.com/cars/2018/10/lone-engineer-spanks-wa...

If a self-driving car can serve for 5 years without a costly overhaul, the amortized cost is only $250k/5 = $50k per annum. Let's say the annual maintenance is $10k, the total would be $60k a year.

It can possibly serve two effective shifts (in fact longer than that but the demand at some hours could be lower and it may need some time to charge, unless battery swapping is performed), which means the cost per shift is only $30k per year which could be in the same ballpark or a bit lower than a human driver would make per shift (depending on the market it serves, etc.)

This is just the beginning though and the costs will most likely come down while the human driver's costs will likely go up.

The cost will go down fast.

What makes you say that? Computers drive a self-driving car, and they've been at scale for a long time.

It might require moving to the Tesla camera-only approach to bring down costs enough to make self-driving cars viable. But that technology is still nowhere near there.

I'm not an expert, but I'd guess that most of the cost is not the physical hardware, but that each one is put together manually by a team of experts. That's a task that seems able to decrease significantly in price.

Costs always come down with mass production. The more self-driving cars you produce, the easier it will be to bring down costs. LIDARs themselves can likely be had for cheaper if purchased in large quantities. There are also multiple companies working on less expensive solid state LIDARs.

Also, if we assume the cost of a self-driving car is 250K, that might seem like a lot, but if you amortize it over 5 years, that's 50K a year. Assuming that a car can bring in a mean profit of $10 per ride, that means it has to do 13.7 rides a day for 5 years to break even. That might seem high, but these cars are not human drivers, they don't need to sleep or take breaks, they could be on the road 20-23 hours a day, leaving some time for refueling and maintenance. I think it would be possible to bring in a profit with $250K self-driving cars and an uber-equivalent price points. It will become even easier when these cars drop to 200K, 150K and 100K, which is certain to happen once the technology spreads.

The car may be ready 24/7, but humans have a nasty habit of all wanting to travel at the same times. In reality the vast majority of trips will happen at rush hour and most cars will sit idle the rest of the day.

It’s brand suicide to not have a car when needed so fleets will be sized for peak load. When you apply queuing theory none of the numbers seem to add up.

The numbers for just driving around and getting paid for the drive might not pencil out. But if you include the real time data about who (for payment purposes), intended destination, origin, and historical trips, that represents data advertisers will pay handsomely for.

If an advertiser tried to entice people to place a tracking device in their cars today for a reward, that would go over like a lead balloon. But Waymo will get people to pay for the tracking data and enjoy the riding convenience, while vacuuming up insights into consumer behavior at granularities that are the story elements of Black Mirror episodes. Advertisers are salivating at sending targeted offers that add pinpoint location and time to the axes they can already select upon.

While a teeny bopper rides, the screen screams out a BOGO offer to buy “the latest Kardashian shoe just 10 meters from your destination, press the Buy It Now button in the next 30 seconds and we’ll have both shoes ready for you in your size for immediate pickup at our entrance including a FREE gift $100 value for true Kardashian collectible fans! BE Kardashian Glamorous Tonight!”

Uber and Lyft have all that data about their customers today, and could monetize the same way. Maybe that's a significant additional revenue stream, but it's orthogonal to whether the cars are self-driving.

We’re just bouncing low energy waves off shit bro, and hitting them with some Fourier transforms. If you’ve got a Russian mathematician WFH in his moms cabin, he’ll get this thing running on a Celeron.

The computers needed to run self-driving cars are not the expensive parts. The sensors are. And the sensors are expensive because there are so few users.

High-precision accelerometers used to be extremely expensive, with the best having prices in the high 5 figures. Then companies started putting (initially really crappy) accelerometers into smartphones, just to know which side is up. Then economies of scale hit and suddenly every smartphone has an accelerometer better than those super-expensive ones, and cost per part is pennies. (And then drones became a thing.)

The same will happen with LIDAR. Nothing about any of the parts of a self-driving car requires it to be super expensive, it's just that things that are made in quantity are cheaper than special snowflake parts.

$250B

twirls mustache

Reply on news.ycombinator.com