Delphi's self-driving car | Hacker News Reader

Delphi's self-driving car

arstechnica.com

50 pointsbane39 comments

When there is a talk about Apple going into self-driving car, there are hundreds of comments. When Musk mentions that Tesla will have self-driving car almost everyone here has an orgasm, but when a company that's well established in automotive industry drives their car from SF to NYC, we can't get even 30 comments in 12 hours?

They didn't do the drive yet, this is just a piece of fluff to fill the time between their initial announcement and the drive (which will start next week).

Actually they are already on the road.

Thats the problem with voting-based sites. The sexy stories get play and everything else, outside of rare oddities, is marginalized. This is fine I suppose if we want to get a certain superficial benefit from this site, but if you want something deeper, often from what I imagine people think of as "boring" companies or "boring" spaces, it just won't work on a site that has this kind of voting mechanism in place.

Geeks like to think of ourselves immune from things like celebrity, hype, PR, etc, but we actually are very vulnerable to it. We play up this "man of logic" persona, but its just a self-compliment. Not too long ago we were telling each other that $FOSS_project would destroy Windows/exchange/office, or Xioami will kill Android as we know it, or $billionare will fix $unsolvable_problem, or Diaspora will destroy Facebook, or that Ouya would destroy mainstream consoles, but it turns out that "boring" often wins as MS, Google, and Sony keep making profits and maintaining marketshare. We believe these things because they are pleasing to believe. If you got your news and tech opinions from slashdot 10 years ago you'd be shocked to even see Microsoft in business. How are we different today than the usually wrong slashdot narrative of yesterday?

The larger issue is that dissenting voices are too easily marginalized. I'm pretty skeptical of the success of Tesla and Oculus for reasons I feel I present in a mostly productive way, but when I mention them its just a ticket to being downvoted, which on here and reddit just limits the visibility of the comment. Fanboy comments get upvoted to mass visibility and become the defining narrative. Those narratives become the norm and, well, here we are.

Poor car companies. Between the possibility of this technology available as retrofits for old cars, the common use case for this as a taxi without the need to own it(and the efficiencies that come with it) ,the possibility of car sharing using this tech(like uber-pool) , and the long life of many current cars - the future of car companies seems hard(maybe after some brief period of big demand).

You say this like the car companies aren't developing self-driving technology too. They are - they just don't need or want to shout about it as they don't want to cannibalise the existing market until it's necessary.

Even if they're the developing the tech, they'll do everything in their power as to not cannibalize themselves , and that will also contribute to their downfall. classic disruption.

And you can see that in the way they talk: they believe that they should aim at partial self-driving(only on highways) while Google targets full self driving(including cities) which is a totally different challenge.

On thing not clear from this article about delphi - is their ride includes city driving ? because self-driving cars could also cannibalize Deplhi, as far as i understand.

This is "the car companies". Car companies are largely systems integrators, buying pieces from companies like delphi and putting them in a nice box.

Delphi was even at one time a part of GM, it was spun out as an independent company in the 1990s (I guess the company and GM division would not entirely overlap, but whatever).

You're assuming there will be mass movement towards self-driving car pools. I don't think that's the case. I think personal car ownership will persist for quite some time, even with self-driving cars. Having a car in your driveway whenever you want is pretty convenient for just about everyone living in a suburb. Even assuming good cost-reductions from pooling, I'm not sure you can completely make the value proposition to offset convenience.

I wouldn't be surprised to learn that self-driving even without pooling is cheaper or not much more expensive than regular driving - and is sure as hell much more convenient. So i would expect suburb people to use it because they are reasonably wealthy.

As for pooling - that remains to be seen with regards to the suburbs - but it's possible that least some of the rides(especially going from/to work) might fit pooling.

It isn't a binary choice. For example you pay for an individual ride to work, and while on the way the system says "would you accept another rider for a $5 discount and an extra 2 minutes in journey time?" So many variations on this theme is possible making maximum utilisation of the fleet of vehicles, and trading off pricing versus convenience (and likely luxury) all the time.

I'm looking forward to "dining cars" where you get to eat a nice meal while in transit.

US cars are on average idle 23 hours a day. Sharing a car between two means roughly halving the fixed costs of ownership, and the more sharing the more reduction in fixed costs. Additionally running a fleet is cheaper because of economies of scale such as maintenance, insurance, charging/fueling. AAA has an infographic showing the biggest costs being depreciation and insurance: http://publicaffairsresources.aaa.biz/wp-content/uploads/201...

A pool also allows other things, such as pricing based on demand - ie being cheaper outside commute hours, spontaneous pooling (eg for a $5 discount can the car go 2 minutes out of its way to pick someone else up?) and being able to bring the right vehicle for each situation (eg a small efficient two seater for a commute, something larger for a family trip to Ikea etc).

Sure some people will continue to outright own their cars, but for most the pooled economics will be compelling, and companies will be able to instantly adjust pricing and convenience. Or to put this another way, how many thousands of dollars a year cheaper do you think it will take for the majority of car ownership to change to on demand usage?

It may also refactor distribution. People may not go to the grocery store, the store will come to them, or not. Just like in the old days, you'd get a morning delivery, and weekly/monthly too for different kind of products. Stores, if they stay 'material based' will look like IKEA, where you walk inside scenes and pick what you want, while something is loading an SDCar (npi) from a factory not too far away.

Damn I feel like a guy in the 60s drawing this http://imgur.com/vmOYZAl

I can't see many ways in which much of anything stays the same. For distribution my guess is that delivery companies will spring up (quite probably the same as the car pool companies during off hours) and deliver for anybody. By that I mean that my favourite Chinese restaurant currently has its own delivery vehicles and personnel, but I am expecting generic delivery companies for any/all business.

They just need lockboxes in the vehicle which the end user unlocks with their phone/pin/whatever. You could even do things like share your location with the service so that deliveries arrive as you get home, whenever that is.

The "city of the future" drawing is interesting. I'm expecting that commercial space will cluster together, optimised around foot traffic, and without vast swathes of asphalt between them, not too different than malls. Residential space is a bit harder to guess - decreasing density will be more manageable, but come at the expense of travel times and delivery fees.

What I like most about the driverless future is just how locally mobile it will let everyone be. No more licenses, no more having to have good vision, no more having to worry about age, intoxication, and even time of day.

If cars become more useful, they will be used more often and car companies will sell more cars not less. Obviously automakers that end up on the wrong side of this trend might struggle, but overall the industry will be larger.

http://en.wikipedia.org/wiki/Jevons_paradox

All I see is that anything big will go dinosaur at one point.

They've reached El Paso, TX, having started in San Francisco:

https://twitter.com/AmyMessano

(Amy Messano is their PR person)

Apparently they racked up a speeding ticket in Arizona. :-)

Reminds me of drivers grunting when their GPS warns of speed limits (I actually checked, my father's one reacted under 1 second from the moment our car passed the sign on a highway).

They share the parking lot of my company, and the cars look pretty cool.

What I don't understand: cars are not encrusted with sensors like Google's self driving cars.

Would love an explaination of their tech to hide sensors.

Didn't Google just bolt a ton of sensors on top of an existing car and called it a day? Also, it seems that only they are using LIDARs - apparently Tesla is doing self-driving cars without them (look at the Model S - do you see any sensor?), and so are some major car manufacturers.

It's true. LIDAR can make the car look bad. If you add LIDAR in places where you can hide them (like the side of the car) vision is seriously restricted. This is why Google places the LIDAR so high up in their cars. The LIDAR is able to give an almost-birds-eye view of the surroundings the higher up it is. Plus Elon doesn't like LIDAR because it bit him when they tried using it for autonomous landing tech at SpaceX.

Still, cameras and radar only get you so far... I'm also wondering how Delphi is doing all this with the tools they have.

Per their site, the Delphi vehicle has "... four short-range radars, three vision-based cameras, six lidars, a localization system, intelligent software algorithms and a full suite of Advanced Drive Assistance Systems."

http://www.delphi.com/delphi-drive

Right, I meant to say, "I wonder how they're getting so much functionality out of LIDAR without mounting it so high?"

Yeah, but they're using radars. Or fixed-pointing LIDARS. They are more discrete and blend more easily into the car.

As for other companies not using LIDARS, I beg to disagree. They do, but they use the far far simpler (and cheaper) variety.

But the Google tech has a broader view of the surrounding area. In the end, it won't matter as much how pretty the sensors look as long as they get the job done.

On the contrary, two cars the same, it will matter more how pretty the sensors are, as people aren't going to purchase an "ugly" car in droves.

I bet you that if the car has better performance and reliability, people won't give a rat's ass how pretty the car looks.

Also, I don't care what my taxi looks like.

> Tesla is doing self-driving cars

Tesla is absolutely not doing self-driving cars at the moment, they're doing limited autopilot (which IIRC already exists to an extant on other high-end sedans e.g. auto-steering using lane detection on highways)

http://spectrum.ieee.org/cars-that-think/transportation/self...

Current Teslas apparently have enough hardware to drive themselves around, but this year they plan to enable only a limited set of features because of safety reasons.

The problem I have with lane detection is that my state's highways don't have the clearest painted lines. And on the road I'd most want to use an auto-drive system on, it's been under construction since the late 80's so the lines are frequently wrong or missing.

Not to mention the effect that heavy rain and snow has on the visibility of the lines. I think it's a fine supplemental system, but should never be a primary guide.

Google's self driving car program was essentially born out of the DARPA Self Driving Challenge where looks weren't in any way a factor. So they started with the best senors possible, and kind of assume that they'd probably get smaller with time.

Delphi, Tesla and all other car manufacturers came at the problem from the starting point that obviously visual sensors where a non-starter. So they started with the smallest sensors possible and kind of assumed that they'd get accurate enough with time

Eventually they'll meet in the middle.

It's funny how fast things have changed since Thrun won. Before that, IIRC, cars wouldn't manage to drive 10% hitting a rock. All of a sudden : Google Cars everywhere.

I know it's amazing. 11 years ago the state of the art car barely made 12 km, before getting stuck on a rock. And that was without having to deal with any traffic.

8 years ago the state of the art car managed to crawl through a 60 mile fake urban course at a speed of just over 20 km/h

Today 'everybody' has a working self driving car handling real life traffic and normal driving speeds.

When I was in college in the early 00's I worked at a Delphi test lab. Even back then they had sensors under development that were pretty advanced especially for the time. Things like automated cruise control that adjusted the cruise speed based on the surrounding traffic. The leap from driver assisting tech to full self driving doesn't seem that big in terms of the sensors themselves and Delphi has been making sensors that go into production automobiles for a long time.

I was hoping with this title that it was about a self-driving car written in Delphi / Pascal. Turns out I was disappointed.

Completely agree about the prejudice against the well established. I am not affiliated with the Delphi project, but work in this space. Not having knowledge of the effectiveness of the Delphi autonomous system, can't speak to that. But, with regards to their mechanical design, this team has demonstrated autonomous capability with a familiar and covert form factor: obscured GPS antennas, LIDAR and cameras. And computing. Not visible in either the rear seat or trunk [1, 2].

Also important to mention that this product is a partnership between Delphi and Ottomatika [3]. Ottomatika is the company that won the 2007 DARPA Urban Challenge from CMU and provides the SLAM and software expertise.

Admittedly, the Google's of the world help with the lobbying and regulatory efforts which eventually expedite the technology transfer to the market for the Delphi's and smaller startups in this space [4].

[1] https://twitter.com/harrymccracken/status/578624426079494145...

[2] https://twitter.com/DelphiAuto/status/580388500291342338

[3] http://ottomatika.com/blog.html

[4] http://www.getcruise.com

Looks like, similarly to the Google car, this is also heavily dependent on detailed maps of roads. The advanced radar-based sensors are pretty slick, but I'm curious about their limitations in terms of the size of the objects they can detect, and how far away.

My car uses some similar technology for automatic cruise control and crash avoidance, but either the hardware or software is primitive, since it occasionally tries to protect me from rear-ending a car in the other lane.

I'm interested in finding out what kind of inter-car communication these companies are considering; direct reports from cars on the scene as well as official beacons that cars can subscribe to along highways and in cities.

Reply on news.ycombinator.com