News for hackers. Stuff that matters.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by toomuchtodo on Sunday January 01, 2017 @11:32PM (#13299049)
Didn't the DARPA grand challenge show self driving was a software problem, not a hardware problem?

Tesla has a level 4 autonomy demonstration video on their site. Maybe computer vision is superior to LIDAR; not a lot of human drivers with laser arrays on them driving around.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by justinpombrio on Sunday January 01, 2017 @11:58PM (#13299142)
> Didn't the DARPA grand challenge show self driving was a software problem, not a hardware problem?

Yes, the software is the hard part. That's why it makes sense to use hardware that makes the job of the software easier. LIDAR tells you how far away objects are; if you're using cameras you have to do that with software instead. AFAIK no one yet knows how to do computer vision with cameras in the sorts of situations a self-driving car needs (though I'm not a vision researcher). That means that Tesla is betting on solving a research problem to get the cars they've already sold to start driving themselves. I always thought that was an insanely audacious business plan.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by jon-wood on Monday January 02, 2017 @04:17AM (#13300030)
To be fair Elon Musk has a track record of coming up with insanely audacious business plans and then following through.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by taneq on Monday January 02, 2017 @01:59AM (#13299538)
> a software problem, not a hardware problem

I don't think this term even means anything. The problem can be solved by various mixtures of hardware plus software to varying degrees of adequacy.

On the one hand, we know that it's possible to drive a vehicle with high reliability in a wide range of circumstances with just two relatively low resolution cameras... if those cameras are feeding into a trained human-level intellect. What combination of hardware and software is required to generate that intellect is still unknown.

On the other hand, the problem seems relatively well solved using 3D laser scanners and relatively straightforward software. It works pretty well and just needs to be shaken down in all of the possible circumstances it might encounter. It's also unreasonably expensive.

We seem to be converging on an "easiest" solution which uses a mixture of cameras, radars, ultrasound and maybe LIDAR, using deep learning for image processing. That's just one solution, though.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by halomru on Monday January 02, 2017 @06:28AM (#13300538)
>with just two relatively low resolution cameras...

We probably can, but most of us drive with two cameras with approximately 120 megapixels of feature resolution with highly optimized non uniform pixel density and incredible dynamic range. Micromovements allow the software to boost the effective resolution by about a factor of four, to over 400 megapixels. Image stabilization works in multiple layers and executes stabilization movements within 10ms in the fastest layer.

The human eyes and vision system are certainly one of our more impressive parts, and we have a really hard time replicating their performance.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by robotresearcher on Monday January 02, 2017 @12:06AM (#13299182)
Here's the winning car from the last GC:

http://www.tartanracing.org

Lots of LIDARs, lots of cameras.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by toomuchtodo on Monday January 02, 2017 @12:34AM (#13299285)
"Stanley", the VW Touareg that won the 2005 grand challenge, also used LIDAR (but fused with camera imagery). In the event that the fused sensor data reduced confidence, the vehicle would slow and lean more heavily on the LIDAR sensors.

"Additional guidance data was provided by a video camera used to observe driving conditions out to eighty meters (beyond the range of the LIDAR) and to ensure room enough for acceleration."

"Stanley was characterized by a machine learning based approach to obstacle detection. Data from the LIDARs was fused with images from the vision system to perform more distant look-ahead. If a path of drivable terrain could not be detected for at least 40 meters in front of the vehicle, speed was decreased and the LIDARs were used to locate a safe passage."

This computer vision processing was performed with 1.6ghz Pentium M processors 11 years ago. Is it so far fetched Nvidia parallel processing hardware can't do better with CCDs than LIDAR?

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by argonaut on Monday January 02, 2017 @02:41AM (#13299643)
This is a research problem, not a compute problem. Directly trying to replicate what LIDAR gives you (SLAM, a 3d map of your current environment) using just cameras is still unsolved today. And it is an active area of research.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by toomuchtodo on Monday January 02, 2017 @11:31AM (#13302093)
I agree it's a research project. Who better to perform said project than Tesla? They have 100k+ vehicles on the road being trained by drivers at a rate of over 2 million miles per day.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by argonaut on Monday January 02, 2017 @10:50PM (#13306538)
Tesla does not have the world's experts in visual SLAM. Those people are in academia, or at Google/Microsoft.

The 100k car/driver thing is a red herring. Steering angle / brake / acceleration info is not the kind of data you need for this task. Big useless data is still useless data, even if it's big.

The right kind of data, if this is to be treated as an ML problem, is a 3D model of the environment (this would be the label), along with camera data (this would be the data). Ironically you need LIDAR to get those labels. It bears noting that the state of the art in visual SLAM is not ML based (it's really difficult to get the "ML" problem I just described to actually work - and it's not for lack of data, Google probably has petabytes of this data from their cars).

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by greglindahl on Monday January 02, 2017 @11:38PM (#13306764)
I thought that the data Tesla is collecting is things like "places where drivers always steer unexpectedly" or "places where drivers always stop but our system didn't detect a stop sign". Those are labels.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by argonaut on Tuesday January 03, 2017 @02:25PM (#13312180)
Those are not the labels for the problem I was talking about (visual SLAM).

Those are labels for a different problem.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by snovv_crash on Monday January 02, 2017 @06:10AM (#13300473)
No, it is solved and there is commercial software that will do this for you. Look at Pix4D, AgiSoft, DroneDeploy, etc. The problem is that it typically doesn't run in real-time, and scales badly with number of images.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by robotresearcher on Monday January 02, 2017 @02:00PM (#13303449)
So it's not a solved problem then. Apart from realtime there is a robustness problem that the offline reconstruction systems don't suffer from too badly. Missing data and poor reconstructions are inconvenient for an offline system but could be deadly in a car.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by snovv_crash on Tuesday January 03, 2017 @06:18AM (#13308399)
Parent said it is a research problem, not a compute problem. I asserted it is a compute problem because we already have robust 3D reconstruction from cameras, but it doesn't run in real-time. If you can come up with a 500GHz CPU with a GB of cache, the problem is solved. It is a compute problem, that we may be able to bypass with research.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by argonaut on Tuesday January 03, 2017 @02:28PM (#13312210)
1) "Scales badly with the number of images" seems like a research problem to me.

2) In theory almost everything is "just" a compute problem (just enumerate or search over all possibilities - of course this may not finish for millenia). But if Moore's law says the compute is 20 years away, I think it's reasonable to call it a research problem.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by agildehaus on Monday January 02, 2017 @01:31AM (#13299459)
The last Grand Challenge was in 2007. A lot has changed since then. Tesla's prototypes are doing what that car did, and quite a bit more, without any LIDAR.

Not to say LIDAR isn't essential. I think LIDAR will get cheap enough that it'll just be silly not to have it.

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by argonaut on Monday January 02, 2017 @02:48AM (#13299666)
Tesla is doing highway driving and some turns without LIDAR. The point of LIDAR is to help you parse a complicated urban scene (and so far what Tesla has are demo videos of urban driving). Of course you don't need LIDAR for the highway.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by sbierwagen on Sunday January 01, 2017 @11:49PM (#13299112)
Animats would indeed be a good person to talk to about CV/LIDAR in the DARPA challenge, since he lead a team that entered it.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by datguacdoh on Monday January 02, 2017 @11:58AM (#13302345)
That demonstration video is very deceiving. There are many times where they switch camera angles and the car has driven some distance that isn't shown. They used clever editing to make it feel like it was 100% autopilot but if you know the area (I live nearby) you can see the jump cuts. I'm hopeful they can do full autonomy with their sensor package but I'm still skeptical at this point.
Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by wyldfire on Sunday January 01, 2017 @11:59PM (#13299150)
I think the safety standard for autonomous car drivers will be much, much higher than human drivers. So it's not fair to compare with just a stereographic camera pair.

IMO the most successful implementations are ones that can predict accidents that humans couldn't even see taking place (obscured by other cars, etc).

Re:Tesla Rolling Out Autopilot Software Updates to 1000 Cars by toomuchtodo on Monday January 02, 2017 @12:03AM (#13299172)
> IMO the most successful implementations are ones that can predict accidents that humans couldn't even see taking place (obscured by other cars, etc).

Such as Tesla'a forward facing radar, highlighted the other day on HN preventing an additional accident participant by detecting a vehicle slowing down ahead that the driver could not visually detect?