Hacker News Reader

Comment by mixedbit | Hacker News Reader

Well, if they expect to use them as self driving cars (which the article mentions), the investment can pay out pretty quickly. Say they can make on average 5$ per ride. To make $60k they need 12k rides. A self driving car should be able to easily make 12k rides in one year (only 32 rides per day).

View on HN

Only 32 rides per day? I would think that's pushing the limit. You will have to buy cars for peak times, so most of them will see only a few hours of use each day. At a guessed 6 hours of busy service a day, that's 5 trips an hour, or 6 minutes to drive to a customer and 6 to drive him where he wants to be.

Also, a risk here is that other large parties will make the same calculation, flood the market (anybody with money will be able to start a cab company with a lot less hassle from labor laws because you don't need that much personnel) and drive margins down.

To win in this space you probably will need both a lean model of operations and a good image in the customer's eye. Uber currently seems to have both, but committing _now_ to a provider of cars is risky (what if other manufacturers turn out to have better or cheaper self-driving cars?). On the other hand, it also may be necessary to commit now in order to stay lean (committing now probably gets them a nice price cut)

Perhaps they're purchasing these to handle the constant/baseline load of riders throughout the day, and then will continue to use human drivers & contractors to supplement demand beyond that, eg. the peaks times you mention.

So it'd be the best of both worlds for Uber: being able to put the self-driving vehicles to use around the clock and not having to make a capital investment to handle spikes in demand, which would just continue to be handled by existing drivers (for now). Just call in the human backup to help out when required.

I'm also wondering if they made the huge order so that they'd be able to influence and guide the requirements for the vehicles to Uber's specs. It seems that the requirements for a self-driving car for an individual's use differs from that for commercial taxiing.

6 minutes to drive to a customer and 6 to drive him where he wants to be.

You're assuming that each ride is independent, but that's already not the case with their Smart Routes - the driver picks up a customer, then picks up another one on the way, then drops off the first and picks up a third, etc.

The goal stated by their CEO is "perpetual rides", in which the car is never without a passenger.

Perpetual rides may be their goal, but I don't see how they will not get there. For example, there's no way they are going to find as many people wanting to go to sport stadiums as are wanting to get away from them, just after the end of the game, and stadiums typically are so far out of about every other destination that they cannot make a detour there with a paying passenger on board to pick up a second passenger.

That applies to a lesser extent to such things as theaters and cinemas, and even suburbs (who wants to get picked up there at 1 AM to fill the car that that theater goer took home?)

They will get below 50:50 for "at least one passenger on board", but it would surprise me if they managed to beat a 25:75 ratio.

>> "...what if other manufacturers turn out to have better or cheaper self-driving cars?"

Cheaper, certainly. Everyone will have cheaper SD cars than Benz, just as everyone has cheaper cabs than an S-class today. But better? If history is any guide, German luxury will be the gold standard of automotive UX for our lifetimes and beyond.

I don't think they need to tackle peak driving time at the start. They still have all of those regular drivers. Surge pricing should still draw out humans.

That was my thought actually - but reversed: Why on earth would they need that amount of cars when self-driving cars are just around the corner?

If these are self-driving cars, it makes sense to me, otherwise it doesnt.

It's hard to come up with a huge number of cars on the spot once self-driving cars are available. It's much easier to retrofit your already existing cars.

Retro-fitting. They buy plain old cars and then slap their sensor suite on top.

That seems way harder unless they plan on buying a very small number of car models. Every different car model (even the same model between years) has different performance characteristics (do you issue the same orders for "hit the brakes" to a car that needs 100ft to go from 40-0 as a car that needs 50ft?) and physical dimensions which would necessitate different software optimizations and sensor calibrations. Development plus rigorous testing of every configuration means this quickly gets out of hand unless you standardize on 1 or a very small number of models.

Hence buying 100k of the same car.

Reply on news.ycombinator.com