Unhappy Truckers and Other Algorithmic Problems | Hacker News Reader

Unhappy Truckers and Other Algorithmic Problems

nautil.us

166 pointsbdon38 comments

Nice article. I would love to know more details about some of the specific examples, though. Like this one:

Yellow Freight used to have some 700 “end of lines,” Powell says, which are sorting terminals where cargo is transferred to its end customers. Powell developed a model that delivered a counterintuitive message: Trucks were traveling farther to get to the customer with so many terminals. Today, he says, Yellow Freight has 400 end of lines. “That was the right number,” he says.

I can understand 400 endpoints being more efficient for some overall process, but I have a hard time seeing how it could reduce miles travelled without some sort of unmentioned complexity (like sending stuff to the wrong endpoints through confusion). Or maybe it includes miles driven taking stuff to the endpoints in the first place? That I could see.

All of this makes me wonder: will business be the first to adapt to self driving vehicles, as opposed to passengers? or are there still significant legal barricades to impede either one now?

Of course, you still need someone to actually make the delivery; even a self-driving car can't really give a person a package.

At least, not yet.

That is true. I guess it just made me wonder since he was going over how much trouble incorporating the human element into the equations is, and maybe trying to do without might be more practical (or maybe more predictable) going forward.

For consumer deliveries at least. However, (I suspect) much of the trucker industry goes to businesses, which would be much more willing to have an employee unload the truck when it arrives.

Truck drivers don't get commercial licenses so they can drive a truck across the country on a highway. They get them to prove that they can back a truck up, turn around tight corners, not run over pedestrians, and most importantly, to be able to navigate an unpredictable, and sometimes rapidly changing world...one where a loading dock might be underground, or it might be around back in that tight alley that doesn't look an inch over 9ft wide...and where your route may go as planned, but it may also require a detour at moments notice because the Highway Patrol is hurredly moving cars off the road so they can set a spike strip to catch a vehicle in a high speed chase.

That isn't to glamorize nor humanize the profession...merely to show the obstacles that truck drivers deal with that are still too fringe to be modeled correctly or trusted in a non-supervised environment. There may be a not-so-distant future where driverless trucks take over long haul drives, with actual drivers handling the last mile. But don't expect the trucker to die quite so quickly.

Watching a 16+ wheeler go down a winding driveway backwards when its only got about 1 foot of space on each side is far more entertaining and impressive than half the events at the Olympics. I could watch this all day.

Are any of those problems substantially more difficult for a computer to handle than they are with a normal car?

Trucks are larger and heavier. That means they take a lot longer to stop, so you have predict things around you further ahead. Also, you can't jackknife a car - I'm guessing the physics of driving a Mac truck with a heavy trailer and 18 wheels is a lot more complex.

But the physics (mass of the load/center of mass of the truck load system, their moments of inertia, N on each of the tires , etc) can be factored into the equations and solved in real time, no?

How far is the line of sight for google self-driving cars?

Shouldn't it be possible to build stations where self-driving trucks can easily stop, and drivers can take over for the "last mile"? This eliminates the dangerous, boring part, allows drivers much more stable lives, but doesn't eliminate them entirely

I wonder if we'll even need trucks. Instead of large trucks carrying lots of goods, what about having more vehicles each carrying less stuff. Something like a cargo van.

I wonder if the reason we use large trucks is so that we can have only one driver and not pay for more. I wonder if having multiple vans instead of a single truck uses a similar amount of fuel (per pound of cargo).

Few reasons for a large truck:

1) The space needed to drive safely behind the car/truck in front of you would be wasted. Cargo, in a container (semi-van, etc.) does not need to travel a safe distance behind the cargo in-front of it; it is tightly packed, and there is a lot of 'cargo'. If all cargo was transported via. cargo vans, the wasted space in-between vehicles (distance for the same volume of cargo traveling with a semi-truck) traveling down the highway would add up to a HUGE distance. Btw. Lets see it in practice:

- The average distance a Truck Load semi-truck travels down the highway is ~856-miles

- DOT Guidelines suggest a three-second rule for driving behind a vehicle. So... at an average speed of 55 m.p.h. A car is traveling at 81ft/second. At three seconds this is 243 ft behind the car in front of you.

- A normal truck container is 48' (we'll go with 48' but there are 53'), and holds 24 pallets. 12 and 12 (side-by side), from front to back.

- I am not sure how many pallets a cargo van could take, but let's go with 2 (it is probably less than this).

At 2 pallets for each cargo van, we would need 12 cargo vans for the equivalent of shipping 1 semi-truck.

At an average of 55 m.p.h. on the highway, 12 cargo vans would total 2,916 ft of wasted space in-between. That's more than a 1/2 mile

* we, keychainlogistics.com have over 40,000 shipments on our system, right now, that would go on a single semi-truck. Most for today. In calculating needing cargo vans for these loads (sure they very in size, but lets go with a total size of 24 pallets per shipment) it would equal 12 * 40,000 = 480,000 vehicles.

With an average of 2,916 ft of wasted space in-between each truckload worth of shipment (calculation referenced above), we are talking a total of 40,000 (combined 12 cargo vans here) * 2,916 ft = 116,640,000 ft of wasted space which = 22,090.91 miles wasted

1.b) Think about the congestion on highways!

2) Large items needing to be shipped: Equipment, Logs, containers, etc.

3) Freight containers from ports. They are a standard size, and the maritime shipping industry, until 1950's or so did not use containers. The book - The Box: How the Shipping Container Made the World Smaller and the World Economy Bigger - gives an amazing/detailed understanding to the efficiency of containers. Which ultimately you need a semi-truck to transport a container.

4) The time to load palletized goods (something that would likely not fit in a cargo van) is much shorter than loading individual boxes

Actually, delivering a package is nothing more then someone who walks in the truck and using a QR-Code (smartphone) as authentication.

Or just let them put a NFC sticker on their smartphone to automate it more..

(Explanation is simple, execution is something else though)

PS. The only problem i really see, is making deliveries in snow (can't detect the road).

> The only problem i really see, is making deliveries in snow (can't detect the road).

Surely they can use sonar/radar to see the road through the snow?

Not really, it's one of the problems (so i've read) Google's automated car is still struggling to handle.

Road, Bicycle part of the road, road marks, ...

How would a pneumatic cannon that launched my package from the truck at the curb be any different from the current driver who slings it from the driveway onto my doorstep?

UPS would no longer be able to deny that this is what happens!

Self driving vehicle still need a conductor. It's not safe enough to have a vehicle that have no one to stop it. Metros, trains and planes have drivers even if they are doing nearly nothing. They are there in case of a problem (which happen too often...).

Lots of trains have no drivers at all: http://en.wikipedia.org/wiki/List_of_driverless_trains

My strong suspicion is that it's going to be closed-loop and business-oriented services that adopt autonomous vehicles first. Deliveries within a factory or campus complex, people-movers, and the like.

Delivery in rural areas where there are long stretches between deliveries (so that loiter times and waiting for someone to come and receive a package isn't a large overhead), and the vehicle could page recipients in advance, skipping those who don't respond, or where deliveries are headed for depots (e.g., local warehouse or distribution center, food/grocery deliveries, etc.).

This was a really quality article. I wish I saw more stuff of this quality posted on here. Definitely checking this site out more.

I agree, this article is really well written, and a lot of their content seems to be very high quality.

However, a word of caution - major funding for this magazine is provided by the Templeton Foundation, which is a right of center religious group. This magazine is very new, so it's hard to say what influence this has on their content, but it seems like a good thing to be cognizant of. After all, trust but verify.

Sources:

http://www.nytimes.com/2013/05/07/science/a-glossy-science-m...

http://www.slate.com/articles/news_and_politics/assessment/1...

The Templeton Foundation also funds a fair bit of fundamental research in theoretical physics and maths. I wouldn't write them off entirely!

The only problem is MIT's news office; "roughly speaking, P is a set of relatively easy problems, NP is a set of incredibly hard problems, and if they’re equal, then a large number of computer science problems that seem to be incredibly hard are actually relatively easy." (It's wrong first because P ⊆ NP and second because it doesn't tell you what's really at stake here.)

I think you could instead say "Roughly speaking, P is a set of easy-looking problems, while NP is a set of problems which can be solved the hard way, by a brute-force search. If they're equal, then every brute-force-able problem is hiding some much easier solution." You could also add a sentence like, "Most computer scientists believe that this is wrong -- that sometimes you can't really improve on a brute-force search. But nobody knows how to prove that negative, it seems to require understanding everything which computers can possibly do."

In one paragraph you can explain the gist of P = NP, I think, to a lay audience. I guess many of them might not know what a brute-force solution is, but you might replace it above with "trying all the combinations" or so. (Hat tip to Scott Aaronson for an essential element of this explanation.)

While I think your definition is apt, I think you vastly overestimate the number of people who grasp the phrase 'brute-force search.'

People understand what "try all the possible combinations"

means.

Do we really have to be vastly overestimating this stuff? What happens if we use the phrase 'brute-force search' and people go look up what it means? Seems like a net-positive to me.

If we reduce everything for the lowest common denominator, then pretty soon nobody has any reason to raise themselves above that.

This is just wrong. Roughly speaking, P is the set of problems that are easy to solve. NP is the set of problems for which it is easy to verify that a solution is correct. There are problems which can be solved by brute force but which are not in NP.

For example, the traveling salesman problem.

TSP is NP-complete, so it is most certainly in NP.

No. The traveling salesman decision problem (is there a circuit of length m) is NP complete. The traveling salesman problem (what is the shortest circuit) is harder. This has been discussed on HN before.

All NP problems are decision problems. When you say "x is in NP", you're implicitly referring to the decision problem. NP-hard does not mean "harder than NP", although that is its colloquial interpretation.

Something doesn't add up here. The article says:

> For each mile saved, per driver, per year, UPS saves $30 million.

Then at the end it says:

> As for UPS, Santilli notes that a driver in Gettysburg, Pa. is now driving nearly 25 miles less per day, from an original route of more than 150 miles down to 126 miles—with the same number of stops.

So they are saving $30 million * 25 * 365 = $274 billion per year. However, UPS' revenue is only $53 billion. What gives? Is this driver in Gettysburg saving a very atypical number of miles, and the average number of miles saved with this system per driver per day is way less? It's a bit disappointing that they left the actual savings out of the article.

> For each mile saved, per driver, per year, UPS saves $30 million.

Err, there's no way a single mile per year costs them $30 million. Looks like the $30m figure is the total savings, unless they're using gold, single-use robots that burn gold to push the gold trucks.

They did say per driver. It's still a high number of course, but I don't see how you can read that differently? UPS has around 400,000 employees, so if 200,000 of those are drivers that number becomes $150 per mile saved. Still way too high, but not single use golden robots high.

I think the goal has to be to read what it doesn't say.

The most sensible interpretation I can come up with is saving 1 route mile for each driver for a year (that is, chopping 365 miles off the work of one driver). That puts the per mile savings right around $0.40.

Alright, so it should have said:

> For each mile saved, per driver, per day, UPS saves $30 million per year.

That would put the total savings at $750 million per year, assuming that Gettysburg driver is typical.

Reply on news.ycombinator.com