Hacker Read top | best | new | newcomments | leaders | about | bookmarklet login

I humbly propose that if you have perfect agents, picking up 5 people with 3 cars is an incredibly simple problem.

It's only difficult because Uber has to deal with humans - will their cell phone be charged? Will they hear the notification on their phones? What if they forgot to indicate that they've taken the rest of the day off? What if they claim to be available but are actually in the restroom at a fast food restaurant? It's supposed to be a 6 minute drive but what if they make a wrong turn? What if they glance quickly at the map of where they're supposed to go but saw it wrong, then muted the turn-by-turn so they could blast some daft punk? Working with human agents is hard.

If you have always-online self-driving cars, then you've already done the hard part. Directing them where to go is trivial.



view as:

Since when was traveling salesman trivial?

I'm only using 5 people with 3 cars as an example. In reality it's hundreds upon hundreds of people and proportionally fewer cars. I understand there are challenges with dealing with humans that will disappear when we switch to robots, but I can't see the API changing very much. The concept of a vehicle network, customers, app, doing clever things to anticipate traffic and customer desires on certain days, all of these things take (I would imagine) years to understand and optimize.

Your claim that operating such a network with humans versus robots, on a logistical (API) level, have "almost zero in common" is rather brash.


Travelling salesman isn't that hard when you 1) scale it down and 2) add in a load of constraints.

Also, the travelling salesman problem is hard because you're looking for the best solution not a good solution. A good one is pretty easy.


Legal | privacy