Hacker News Reader

Comment by viraptor | Hacker News Reader

Looking at truck size, this doesn't even look like a very hard problem. You can essentially solve it per neighbourhood / post code and then inside each of those sections separately for a good approximation. The interesting question is: why doesn't UPS think that's worth doing? Or are there some fun constraints we're not aware of?

Or specifically: why don't drivers get a preferred route along with their packages?

View on HN

It sounds like they do, but that it breaks down when the trucks get overfull. I wonder if there's a breaking point where the projection of just how full the trucks will be, which packages will get to be thrown as overfill in which trucks, and how many will fit in each, is outside the operating bounds of their system and they know it's only a handful of days a year so they just brute force it instead of investing a lot (more trucks or more drivers would be the non-algorithmic sort of investment) for a recurring-but-tiny edge case.

Maybe it's like a tech company with a known datacenter footprint and standard load pattern, so normal operations can be done in-house cheaper then in the cloud, but twice a year they have to do 200% traffic and just bite the bullet and scale that out in AWS at a much higher price, but only for those two days, compared to capacity that would sit idle the entire rest of the year.

>Maybe it's like a tech company with a known datacenter footprint and standard load pattern, so normal operations can be done in-house cheaper then in the cloud, but twice a year they have to do 200% traffic and just bite the bullet and scale that out in AWS at a much higher price, but only for those two days, compared to capacity that would sit idle the entire rest of the year.

The reason why that mostly works on AWS is that amazon has someone else to sell some of the peak capacity to the rest of the year. With package delivery, you'd need to figure out something profitable to do with the 'holiday delivery trucks' the rest of the year.

Sounds like the rest of the year they handle the accounting...

This is done, drivers do get a route (EDD trace). Problem is that there's a min wage human (machines are too slow/expensive) with heavy time constraints who loads the truck. If they're part of the 1/3 that is a peak seasonal, don't care, or are pushed too hard by management, the load will be bad. Loading stop for stop is hard and has no incentive.

Most routes (at least here) have business stops which make up the majority of the bulk down the middle. If it's a resi route with a lot of irregulars (70-150lb packages, mattresses, etc) it'll be a mess.

The options to solve this are pretty much more trucks (there are none, and nowhere to put them after the "normal" seasonal increase), more drivers (hiring/training is an issue), making multiple trips (depends on distance from hub), or having parts of the route shuttled out.

Just give the incentive. Give bonuses to loaders when trucks deliver more efficiently. It would probably help with turnover, so the peaks would be much better with second nature efficiency ops at the dock.

I would go so far as to say that if they had invested in efficiency this way they could have developed their ORION project for far less money since they would be crowdsourcing at least the first major part of the solution from their workforce. Study that with sensors and data, and develop your algorithm from there. AFAIK, an experienced taxi driver is every bit as fast as a google maps traffic optimized route.

>Looking at truck size, this doesn't even look like a very hard problem

Can we now add UPS to the growing list of other HN-Commenter-Weekend-Projects? So far I’ve got Facebook, Twitter, Shazam.

I think you took the comment the wrong way. I'm not saying UPS is trivially solvable as a company. I explicitly said there's probably some reason why that solution is not used / not working over Christmas. I'm saying the approximation for an isolated "optimised route for these packages problem" for a truck load of packages is not hard - so what are we missing?

When I worked in UPS operations, Peak season (generally starting in August / September when retail stores started stocking up, and building to full force in December) saw volume approximately double.

It's pretty hard to fine-tune a system for maximum efficiency most of the time, then make everything go 2X during a spike with the same level of efficiency. Especially when you're dealing with physical space and people who require training.

Drivers obviously prefer a fixed route they know, but a complex routing algorithm could shift borders of neighborhoods or even interleave routes to meet timely appointments or size and weight constraints. Consider fairness, too, to equalize load.

Overall it seems the problem is fixed with brute force, taken to the extreme as in the topical article.

Reply on news.ycombinator.com