Hacker News Reader

Comment by MengerSponge | Hacker News Reader

Nine financial quarters. Yeah, not happening.

The gulf between "testing fully driverless systems" on desert interstates and "deploying a useful autonomous vehicle" is wiiiide.

My tongue-in-cheek prediction: we'll have commercially viable self-driving cars one year after Amazon warehouses are fully automated.

View on HN

9 financial quarters plus all the prior years of work.

Aptiv appears to have been working on self-driving for quite some time.

https://www.forbes.com/sites/alanohnsman/2019/05/31/self-dri...

https://www.aptiv.com/media/article/aptiv-launches-fleet-of-...

In fact the self-driving part goes back several years:

https://www.forbes.com/sites/joecornell/2017/12/06/delphi-au...

https://www.theverge.com/2017/10/24/16533534/delphi-nutonomy...

Yes, Ottomatika (self-driving car CMU spinoff) was purchased in 2015 and NuTonomy (MIT spinoff) in 2017. They have been busy.

Is anyone actually testing self driving cars on desert interstates yet? From what I understand, most of the testing happens on residential streets in the Bay Area. Testing on desert interstates would be a huge step up.

https://www.theverge.com/2019/8/15/20805994/ups-self-driving...

There's a video out there of the UPS truck driving 90 minutes end-to-end without interruption by the driver or engineer who are along for the ride.

Desert interstates are actually easier than residential streets. No pedestrians, no snow, little rain, no intersections.

High speeds, winding roads that go up and down mountains. It actually isn’t that easy out here.

It should be predictable difficulty though, right? What's the actual difficulty for a robot driver? Typically in residential driving it comes down to perception(/prediction) difficulty and planning difficulty.

Interstate highways have minimum curve radii and maximum grades and a whole host of other requirements that make them easy by definition.

It's not hard to do. Google has been able to do it for some years. It's difficult to build a service around cars that are only autonomous on the highway, so the autonomy companies pivoted to in-city driving as a natural next step toward having no safety driver.

[edit: it's hard to do, but has been done a couple times over by a couple of companies]

I've been waiting for someone to try Montana in the winter. I'll believe the self driving thing when that works. Might be able to save some lives and keep the number of roadside crosses down.

Having done SLC I15 to Spokane I90 (the route that takes you into Idaho twice, and crosses the inter continental divide a few hundred times) in the winter before, I’m not even sure human-driving works well across those passes. Was in first gear going down while locals were passing me at 80 MPH.

Canadians[1]! A whole country, full of innovation north of Montana and frighteningly aware of snow and ice :).

[1]: https://www.youtube.com/watch?v=7zaf4pSe4v4

Yeah, but Montana is, well, Montana. I'm in North Dakota and we have some really scary driving weather including the ice rink known as I-94 near Fargo. But, Montana has such a variety of driving conditions including mountain passes from heck, and a tradition of driving a bit fast to make life difficult for any automated system.

Also rain. LIDARs don't work in the rain, and driving in rain is quite challenging in urban areas.

Hello from Michigan, and ditto. I've been dodging potholes that could eat an entire motorcycle, and I have yet to see a so-called "self-driving" car that even understood the concept of a pothole.

Or random road construction that just closes an entire lane on a 1-lane road, so you have to divert into the turn lane. No arrows, just cones, figure it out. This is obvious and natural for human drivers...

Or flagmen at a construction site, with the "stop/slow" sign....

Reply on news.ycombinator.com