[r3m6]

Is Lidar really required? I assume Google uses it only because it does not care about cost at this point of the development.

German car makes (Mercedes, Audi, BMW,...) think they can create a driverless car without the need for expensive Lidar, see for example http://spectrum.ieee.org/cars-that-think/transportation/self...

[ghaff]

Although that article also makes it unclear the degree of autonomy we're talking about here, using Lidar or otherwise. The big challenge, as noted in the link) is that there's this space between drivers still needing to drive (with assistance) and true driverless 99.999% just works. And within that gap, you potentially don't want to introduce certain automation because it will make drivers inattentive even though the machinery isn't quite up to the task of handling every contingency on its own. The billion dollar question is the timeframe for the 99.999% case--which enables a whole lot of radical use cases which the 99% case does not (such as no driver or no attentive driver). Just because the 99% use case could be here relatively soon doesn't mean that the 99.999% case couldn't be much further off (as in many decades).

[bentcorner]

> which enables a whole lot of radical use cases which the 99% case does not (such as no driver or no attentive driver)

I'm excited for this. Car ownership and operation could be an interesting historical note 200 years from now. Family scheduling would become much more flexible and children incapable of operating a vehicle will be empowered to travel long distances.

[ghaff]

I'm not convinced about car ownership. There are a lot of benefits in owning a car that you can keep "stuff" in. But just as Zipcars have presumably eliminated some car ownership use cases at the margins, fully autonomous cars would have a significant effect.

[bentcorner]

> But just as Zipcars have presumably eliminated some car ownership use cases at the margins, fully autonomous cars would have a significant effect.

The big big thing I see about driverless cars is:

a) You can use it on-demand, even more easily than Zipcars. No need to physically be located near a lot, I can "hail" a driverless car and have it waiting for me when I need it. Smart routing and allocation can ensure that capacity is used smartly - it could drop somebody off while picking me up.

b) People who can't or won't drive can benefit from this. From young children to the elderly, transportation becomes much much easier. Have to take your kid to an appointment in the middle of the day? A driverless car can pick them up from school and bring them directly to the appointment and you can meet them there. You go to the appointment together, then driverless cars can take each of you back to your daily work, saving time all-around.

Yes - it's useful to have a car to keep stuff in. I can imagine you could keep one "on call" for a day out, and you can meet the car (at whatever exit is closest to you!) throughout the day to get supplies or drop things off.

Maybe I'm being too optimistic about what driverless cars can give us. Realistically I think hands-off driverless cars are within the next few years, but FUD and lawsuits will prevent more radical forward progress. It's certainly not unimaginable that it will take an entire generation of people to die off before truly autonomous cars are common and accepted.

[sremani]

Car ownership on itself may have some benefits but most of the people who own car is not by choice. US cities are planned around cars. Walking and Biking are not only out of option for many, they are actually unsafe once you leave the neighborhood streets.

[dragonwriter]

> There are a lot of benefits in owning a car that you can keep "stuff" in.

So, combine driverless cars with depots with loading/unloading robots, then you can leave stuff in your driverless car, and have access to it with the next driverless pickup.

[bdamm]

Security is an interesting aspect as well. What's to stop someone from planting a location tracker on a shared car, then learning the locations of subsequent users? Or a bomb?

[joshuahedlund]

What's to stop someone from doing those things on cars now?

[krapp]

The most likely novel vulnerabilities with these will be in the way they network with one another and municipalities. Also, doubtless people will try to carjack them or steal the hubcaps and stuff.

[Thrymr]

Elon Musk thinks autopilot will come before driverless, too: http://www.bloomberg.com/news/2013-05-07/tesla-ceo-talking-w... (or at least he did last year). I tend to agree. More cars have things like lane detection and adaptive cruise control that are steps towards autopilot. Truly driverless is not just a hard engineering problem, it is a huge unsolved social and legal problem.

[Retric]

One option for a 95+% driver-less cars is simply having a remote driver that handles a fleet of them.

[ghaff]

Depends on the cases that aren't reliably handled by the computer. "Last 100 feet" problems associated with, say, parking could potentially be handled remotely. The bigger issue though is circumstances that require quick decision making by a competent human.

[cdash]

Would Lidar really be all that expensive if it was used in mass produced driverless cars?

[ddeck]

I'd be interested to hear from any experts in the are on this.

Why is Lidar so expensive? My vacuum cleaner has Lidar. Clearly not of the same grade as that on the Google car, but what is it that makes their unit cost tends of thousands of dollars and is this likely to fall dramatically at scale?

[benjamincburns]

I don't have much direct experience with lidar, but with my background in in situ sensing I could probably land a job working with/on it. If I may be so bold as to presume that I just barely qualify as an "expert" here:

The lidar in your vacuum (presumably a neato-XV series; if so, you have fine taste sir/madam) is no more than a spinning laser range finder. These work by offsetting a laser from a linear image sensor (imagine a camera with a crazy wide aspect ratio, maybe 256x6 resolution) and calculating distance by simple trigonometry or a pre-calibrated look-up table. These are called spinning parallax lidar. Their output is two dimensional in polar coordinates (angle, distance).

More expensive lidar systems use carrier wave phase measurements (modulate the laser to a known signal, look for its reflection and calculate phase difference), direct time of flight measurements (like narrow beam sonic range finding, but with light, requires very expensive timing circuitry), doppler shift, and/or various structured light transformation/interference effects. Some units only use one of these techniques, some use a combination of these techniques. Most of these units operate using a number of lasers simultaneously to spit out 3 dimensional data, either as planar depth maps, or as 3D polar coordinates (angle-x, angle-y, radius). Either way, all of these are expensive because they require very fast and very precise complex analog circuitry, tight controls on noise sources (RF interference, temperature dependence, very good power isolation), tight controls on spurious emissions (GHz oscillators), careful calibration, and most importantly, very complicated software.

To me, time of flight is the most interesting, but the least practical/likely to be made inexpensive. For a DIY example of the kind of timing circuitry involved, check out [1].

That said, there's room to make it cheaper. One such way may be to use a structured light transformation technique similar to Kinect, but speed up the camera a bit and modulate the output to a known signal (gives better interference rejection).

1: http://piotr.nikiel.info/?page_id=95

[eru]

Do the challenges compare at all to what we've seen in solid state gyroscopes / accelerometers getting massively cheaper?

[morley]

Would anyone mass-produce a driverless car if Lidar were so expensive?

[randyrand]

Your questions both have the same answer (ie it's the same question) but that's a backwards way of thinking about it. You have to consider the cost of lidar at economics of scale, not before.

If your point is that lidar won't be cheap enough even at economies of scale to make a mass market device, then just answer OPs question directly. Your comment is just a restatement in a less intuitive way.

[jbarrow]

I actually visited a German University (TU-Braunschweig) that works with a German car company to research driverless cars. The project that I saw [1] used Lidar as a verification method, but the ultimate aim was to be able to drive without it.

The interesting takeaway that I had from the visit was that the team working on this car believes that even partially autonomous cars are a technology that will come into the main stream only several decades into the future.

[1] http://www.spiegel.de/video/fahrerloses-auto-leonie-kreuzt-d...

Sorry, the link is to a video in German. I'm sure that if you search for Braunschweig "Leonie" you may find a source in English.

[seanflyon]

"only several decades into the future" sounds like a conservative estimate for partially autonomous cars becoming mainstream.

[warfangle]

I wonder if they're using Lidar on their test vehicles in order to improve stereoscopic image analysis to the point where they can use cameras instead.

[jre]

As far as I know, stereo vision has some problems with uniform surfaces (snow-covered road). Since any two points on a uniform surface will have similar descriptors, it's pretty hard to match them.

There are some solutions to this problem (using a Markov Random field or some assumptions about the surface), but I'm not sure how reliable they are.

[warfangle]

How do we do it, then?

ed. I guess that's the trillion dollar question...

[trhway]

we have pretty high retina resolution. To feel the limits of our stereo machinery, look at any monotone surface with monotone lightning at the distance that you can' distinguish any local mini-features of the surface - looking at the smooth ceiling of something like this - and you can feel how your eyes strain trying to find (and can't) the correct focus distance to distinguish the features and thus to get stereo working by matching the features from the left and right channels. The same feeling can be felt when you look at say vertical bar patterned surface so that eyes/brain have harder time to match specific bars from left and right eye. In both cases the surface should be large enough to cover the focus spot in your vision space, so that eyes/brain couldn't get help from relative position of the surface vs. other objects.

Those are actually the same issues one encounters when develops computer stereo :) The modern cameras got to the retina resolution and computer power today is able to do stereo match on those resolutions - this is why we started to get meaningful results. The computer stereo will surpass human's because cameras can have higher resolution, higher sensitivity and they can see in different wavelengths in addition to visual. Plus you can "paint" the objects ahead of you with IR for example, so the stereo match would be even easier.

[warfangle]

But with e.g. smooth ceilings, we use other hints to figure out distances - like vanishing points, etc.

[trhway]

this is why it should be pretty large surface with very monotone lighting, not very bright, like at the end of day.

[jre]

In addition to trhway's answer, I'd say we use a lot of assumptions about the world : houses usually have flat, orthogonal walls, streets are flat, etc.. Of course, we can have the computer learn those assumption, but it's more complicated than pure dense stereo.

[Symmetry]

I'm curious how you would detect debris on the road at speed reliably without it. I wouldn't trust a visual system to distinguish obstacles from discolorations, and other systems wouldn't have the precision or range.

[vonmoltke]

It can be done with an RF system using SAR techniques, if the vehicle has a precision INS installed. Fuse that with data from other sensors, and you can build a pretty good topo map of the road ahead of the vehicle.

[zo1]

I genuinely don't know all three acronyms you used there.

Be honest. Do you think that most of the people that are browsing over and reading these comments are knowledgeable enough in your niche field to know all those acronyms?

Probably not.

[vonmoltke]

SAR -> Synthetic Aperture Radar INS -> Inertial Navigation System RF -> Radio Frequency

Yes, I should have spelled out the first two. The third I do expect most people here to know, given the context of the thread.

> Do you think that most of the people that are browsing over and reading these comments are knowledgeable enough in your niche field to know all those acronyms?

RF and sensors may be a portion of electrical engineering that is rarely discussed here, but it is hardly "niche".

I made a quick post where I should have made a more detailed one. Mea culpa.

[jey]

I think SAR = https://en.wikipedia.org/wiki/Specific_absorption_rate

[cdash]

Probably Synthetic Aperature Radar - http://en.wikipedia.org/wiki/Synthetic_aperture_radar

RF is probably radio frequency

INS is probably inertial navigation system - http://en.wikipedia.org/wiki/Inertial_navigation_system

[StavrosK]

Which three? I only see SAR and INS, which I also don't know.

[zo1]

"RF" was the third one. And I suppose I could take a guess and say it means "Radio Frequency". But just because the acronym's letters fit the phrase, doesn't mean I'm right as I am not knowledgeable in the field being discussed.

[robryk]

Are there any inertial navigation systems that work at all in ground vehicles? I'd expect otherwise due to e.g. vibration.