Hacker News Reader

Comment by MontyCarloHall | Hacker News Reader

It’s not a good harbinger for fully unsupervised autonomous driving if autonomous trucking companies are experiencing widespread trouble [0,1,2]. In theory, autonomous trucks should be the first fully self-driving vehicles to see widespread adoption, since highway driving is among the easiest, most straightforward forms of driving there is. If investors are not convinced that these trucking companies are close to nailing that, it does not speak well for the viability of unsupervised autonomous driving overall.

[0] https://www.freightwaves.com/news/reports-autonomous-platoon...

[1] https://www.dallasnews.com/business/jobs/2023/02/06/two-comp...

[2] https://www.truckinginfo.com/10189392/autonomous-truck-compa...

View on HN

How sure are you that driving giant big rigs at 70mph is easier than little sedans at 25mph? Yes highways generally have less obstacles and unexpected scenarios than city streets but not zero, and you still have to deal with them. Doing a safe emergency swerve or stop with a big rig moving 70mph is a LOT harder than with a sedan moving 25mph on a street in SF

One argument might be that the statistics of accidents per hour driven demonstrates that highway driving is easier.

I don’t think the comment is “driving a truck is easier” but that trucks find themselves on the highway more.

Statistically, more accidents occur in highways than roadways. The general idea is that accidents are more prone to happen there's overload of tasks (i.e. changing lanes while maintaining safe distance between cars in multiple lanes).

I'd say driving along a relatively straight road on a high way is much easier (and less accident prone) than driving on a street in say SF or NYC.

Also, will AI trucks proactively move to the left when approaching a short on-ramp to avoid slower incoming cars, that maybe can’t even be seen yet because the on-ramp is curved?

We’re they just trying to automate highway driving though? Driving a 18 wheeler in a city seems like an order of magnitude harder than driving a car in a similar situation. They have to be super careful about being in the right lane ahead of time and to avoid hitting bridges etc or taking illegal routes.

Automate highway driving, then when you get into the city have a real driver (connected via low latency link) drive the rest of the way.

Around here the most horrific accidents are with big rig trucks. These are the last vehicles that need to get confused by a construction zone and jump a lane or suddenly brake.

Autonomous highway driving is not easier than urban, if anything for the typical Lidar-based approaches it's much harder due to limited Lidar range and very long stopping distance requirements for trucks, plus the fact that a bigrig can cause huge amounts of damage in any accident.

The various pure-play L4 companies that pivoted to trucking due to its alleged relative ease were simply exploiting layperson investor bias to rake it in (or possibly even had a similar bias due to lack of expertise). It's an open secret that Embark had virtually no chance to go to market, even in a good macro environment.

Makes you wonder why autonomous trucking isn't the full and dedicated focus of Google et al. Vastly easier to accomplish, and arguably a bigger societal impact than autonomous passenger vehicles.

Plus it just makes sense to solve an easier problem first and work it into the broader solution later

It’s possible they are asleep at the wheel

Curious why you think trucking would have a bigger societal impact? As to why Google et al. are aiming for passenger vehicles, there about 4 million semi trucks in the US and over 290 million passenger vehicles.

And whats the economic value of 1 mile driven per truck vs passenger vehicle?

Cheaper truck transport also opens up more cost competitive manufacturing, factory placement. All of these will be highly deflationary.

In a remote work world a large percentage of the workforce only has to travel for discretionary, not economic reasons.

For a venture, you want the value that you can capture to drive direction, not all the externalities and I think with trucking that's small.

It makes sense for the US Gov, just not for Google.

Waymo is doing trucking too. It's probably a 98% overlap with the skills their driver (the Waymo software) needs for driving passenger vehicles. Automating driving is so difficult, it's a mystery why anyone thought all but the largest, most advanced, and best funded software companies could succeed.

Because maybe it's not vastly easier to accomplish

With tons more liability per accident...

It feels ikea they should start with rail and subways not trucks. If we cant even automate things that are on tracks how can we automate things that aren't?

At least from my recollection, commercial rail has an incredibly high amount of regulation and union involvement. Which means a minimum 2 person onboard requirement. I can’t imagine it being an easy market to break into, despite all the technology being there. Wabtec, the locomotive giant, today has TripOptimizer which is smart cruise control, which might be all you really need for most situations. Barring release of sand and operation in the rail yard.

Commercial trucking is largely the same (safety rules and unions)

The problem is not union or regulation. The problem is that American rail infrastructure is a complete disaster and then train companies use that infrastructure to push incredibly long trains on them driven by crews that have horrible working experience (crew switches in the middle of nowhere in the middle of the night and so on).

Having 1 driver is quite common in most of the world but most places don't run on that kind of infrastructure with that kind of loads. 2 crew has prevented massive desasters in US rail history.

If anybody was serious about logic, upgrading the rail infrastructure, totally new signaling system, electrification, shorter trains would allow for a huge amount of self driving.

But don't expect that rail companies will do any of that. They have very little punishment for there trains falling over, or delay caused by Amtrek and they can do with their crews as they like.

Could it be, that just for the potential of a malfunction or accident, it’s beneficial to have a human on board to handle such situations?

And if that’s the case, the human might as well just drive the train while he’s waiting.

The same is true for tractor trailer trucks.

Of course we can automate trains. I'm working in a research project doing that right now on a rail track between the Netherlands and Germany. It's quite a bit of work but I don't see any fundamental issues.

Straightforward enough:

https://www.youtube.com/watch?v=hf5iBbmas2w

... until it isn't:

https://www.youtube.com/watch?v=Ec8Yx-Mp-5s

Linked more for your interest than to make any point - the derailmant issues in the second link weren't directly linked to automation, but we do have big train sets here in W.Australia.

It is for humans, but I don't think computer vision is where it needs to be to make it easy for computers.

Lower level semi-autonomous driving aids like Tesla's FSD can effectively use side channels (eg other vehicles) in some situations, but larger vehicles aren't able to do this as effectively.

Last but not least, there are tons of edge cases still handled by humans even if the above 2 points were handled.

I think the norm will be semi-autonomous low level driving aids in many conditions to more autonomous higher level driving aids in more limited conditions, but not fully autonomous vehicles in many conditions.

Long haul trucks don’t just drive on highways. They need to navigate surface streets, along with refueling and weigh stations.

Reply on news.ycombinator.com