Hacker News Reader

Comment by ryannevius | Hacker News Reader

I don't mean to be a downer...but is there really anything novel here? I understand that this is a first step, but I feel like Uber could have significantly reduced the manual labor required (e.g. to disconnect/connect a trailer). It seems that the load (mental/physical) on the "driver" of the self-driving semi will be nearly equal to a driver who has cruise control.

View on HN

I am not sure how much of that equipment is legislated. The air fittings for example are of a type I have never really seen elsewhere. They have been standard in the trucking industry as long as I have known (Oldest truck I have been around was from the 70's)

While you could develop a way to robotically latch those it seems like it would be easier to redesign the fitting.

You’re right it’s not that novel. Volvo has self-driving* truck convoys going back to the 90s I think.

I think for long-haul trucking it’s certainly pretty hard to stay fully tuned while driving for such long times. And for truck companies, they can save tons of money on labor when they don’t have to pay the driver for the hours on the road and just for the manual disconnect/connect docking. Maybe they can make it fully automated, but why haven’t any trucks done that already? It’s an easier problem to solve than self-driving.

* by convoy, I mean that one truck up front was driven by a human and then a few were automated to draft from directly behind to just follow the truck in front of them.

EDIT: Here's a source for the 90s, where this paper had a few authors from Volvo.

from: http://orfe.princeton.edu/~alaink/SmartDrivingCars/ITFVHA15/...

>Additionally, different applications of automated driving for trucks have been demonstrated across the globe since the 1990’s. The focus has been on proof of concept for truck platooning due to the foreseen fuel economy and traffic flow benefits.

Since the 90s?? That's very impressive if true. I remember reading about a proposition to do this for cars years ago, before any talk of self driving vehicles. A car would lock in behind a professional driver vehicle such as a truck or bus and match the turning and braking. I didn't think the tech existed though

The 90s wasn't exactly the stone age, there was plenty of self-driving vehicle work taking place. For example:

https://www.smithsonianmag.com/history/the-national-automate...

I edited the original comment to include a source. Serious talk of self-driving cars was around by then, as they had already successfully used auto-pilot in planes, it wasn't that infeasible.

It seems like truck convoys are a less efficient reinvention of trains. I'm still surprised they didn't take off in the States, though, since our rail infrastructure is pathetic. Maybe it only works now because Uber can burn $1B a quarter.

Why do you think our rail infrastructure is pathetic? It's the largest freight rail networks in the world.

It sucks for passengers but that's not relevant to this discussion.

You're right -- it's better than I thought, though it's still in the middle of the pack when measured by track per person or track per area:

https://en.wikipedia.org/wiki/List_of_countries_by_rail_tran...

I guess I'm biased by seeing all the abandoned rail, since it's well down from its peak.

> it's still in the middle of the pack when measured by track per person or track per area

Both of those seem like somewhat irrelevant metrics. Why should a more densely populated country be penalized for having larger clusters of people compared to (based on this data) Canada or Russia? On the other end, why should a country with large swaths of unpopulated land cover it with rail to improve its track per area metric, which densely populated countries seem to do well on?

It may be more useful to consider the efficiency and cost of freight rail, as well as the share of freight transit (per unit of distance, for instance) that is on rail.

They're not perfect, but they seem useful to me: roughly how much stuff per mile is on your rails, and how far do you have to travel to reach it? Do you have some cost per mile figures to share?

Check out the usage statistics.

https://en.wikipedia.org/wiki/List_of_countries_by_rail_usag...

Most tonnage shipped by rail in the world, with 10x more than the EU.

Yeah, the US rail system is the best in the world for freight. In other countries, passenger rail has priority, so the freight service suffers a bit. In the US, it's the other way around.

That's also why Amtrak is such garbage compared to any civilized country.

Truck companies avoid this kind of thing because it's expensive to maintain pools of short-haul drivers at either end of the transport link. If you are willing to assume those costs/risks, you can already send freight by rail, which is an order of magnitude more efficient on fuel, and sometimes even a little faster.

The reason you pay for long-haul trucking is that you want to have the driver _already on hand_ for the two short-haul trips at either end of the journey, navigating complex issues and handling cargo loading. The downside is that you have to pay the driver for the over-the-road portion. The upside is that there's zero chance of a driver not being available, and zero chance of the driver being left idle waiting for the long-haul transport to arrive.

The only novelty here is that Uber is the one paying the pools of short-haul truckers on either end, and assuming the risk that they will sit idle. (They are, of course, not going to tell us if it is unprofitable!)

The US already has one of the largest freight rail networks in the world (at the expense of having a totally shitty passenger rail network; although that's starting to change with Florida and California's efforts).

I wish more effort would be put into expanding freight even more so you could get the same speed and efficiency for shorter hauls, mail and package delivery. It seems like that would be way more worthwhile than automated trucks. It's substantially more trivial to automate freight trains as well. Passenger trains in Singapore, Malaysia and parts of London are already fully autonomous with no driver at all.

The U.S. used to have extensive short-haul rail, even inside cities. That's how cargo moved around back in the era before inexpensive motorized trucks.

I suspect the reason it disappeared is that rail sidings are large and expensive compared to truck loading docks. You're not going to have ten sidings on a property as easily as you stick ten bays in a loading dock.

> Truck companies avoid this kind of thing because it's expensive to maintain pools of short-haul drivers at either end of the transport link. If you are willing to assume those costs/risks, you can already send freight by rail, which is an order of magnitude more efficient on fuel, and sometimes even a little faster.

Speaking only from the US perspective, I agree with most of this. But there are a couple things you're not taking into account:

1) Delivery variability for rail. While rail is much cheaper in terms of cost per ton mile, good luck getting an accurate estimate of when it will be delivered. Midwest to Pacific Northwest I've seen variability of a week or two for going halfway across the country. If they're delivering to you and are multiple days late, they still expect you to drop whatever else you had planned and get their rail car unloaded on a relatively short turnaround or they start charging fees.

2) Rail requires quite a large scale. Many factories aren't shipping out more than a few semis per day. Getting to enough to fill a couple rail cars would significantly increase the inventory levels.

My guess is the point of this video is to paint a picture of AI as really pretty boring and nothing that much different for the driver.

This is a tactic to avoid pushback from the public / truck drivers.

This is an attempt to control the narrative on what autonomous shipping is going to look like.

Reply on news.ycombinator.com