Quoted:
  I'm talking about implementation, not technical ability or sophistication. The reason your Garmin sucks? Not technical, it's the failure to build accurate maps and maintain them. That's essential for autonomy. I navigate by GPS routinely, and there isn't a solution on the market that doesn't occasionally just fucking suck....but this is going to be the foundation for how these vehicles navigate? Recipe for issues.

Quoted:


"I'm sorry but you have exceeded your carbon credit allowance for the month. Vehicle will remain parked until credit balance resets to zero in 14 days. Would you like the location of the nearest mass transit station?"

"Today is a red air quality day. All leisure travel will be limited."

"A shelter in place order has been issued. All vehicles will remain disabled until the order has been lifted."

Nothing good will come of this once the goverment gets involved and you know they will. The possibilities of fuckery are endless and that's not including cyber attacks from foreign nations, terror groups or bored script kiddies that could wreak havoc on the transportation network. Fuck self driving cars.

Quoted:



Lol, meanwhile planes are landing themselves on aircraft carriers.

Such technology. Much scare

Quoted:
I'm sticking with my horse and buggy. My horse Roach knows the way home and once I point her in the right direction she'll take me home. I can nap in the back without worry. She's OK with coyotes and rattlesnakes but doesn't do well with stallions when she's in heat.



/this

I've been hit by exploding re-treads seven times this week and it's only Tuesday.



People in this thread failed to mention the number of times jumbo jets land on freeways, boats tear loose from trailers, some old guy has a heart attack and plows though the liquor store, and the ever present sharknados in sharknado alley.

Quoted:


No, it is not. Not even remotely.  



I talked to this. Interface layers designed for extensibility are the reasons you see this. That is not/rarely the case in hardware specific implementation. If it was, right now you would be having to disconnect your car's battery every other week because the ECU was borked. "Autonomy" is implemented no differently than how your injector duty cycles as they correlate to spark advance/retard and air temp. Either lookup tables or hard coded formulas. This is how it is actually done in "AI," gains determined through some type of optimization training. Once you have those, it can be hard coded. What looks like "complex reasoning" or "decision making," really isn't.



Depends on the manufacturer.



From the outset, no. Just like anything else. So how and to what degree they implement it will be critical to adoption. I believe we are already seeing the phased smart approach right now. Auto parking, adaptive cruise control, lane departure warning, collision avoidance assist. Just gradually keep on rolling in new features until eventual you've covered most of the decision space.

I absolutely agree, lawyers could easily kill this.  But if auto manufacturers are smart, they will never present it as "autonomous" car. The next thing you will see should be "adaptive cruise control with lane departure warning" two things already out there. Then after that "adaptive cruise control with lane departure warning and correction." etc etc. Just keep rolling in more.

Incrementally... the same way the left is eroding our gun rights.

Quoted:

  Sounds like an AGW guy screaming "DENIER!!!" at the skeptics.

Quoted:
No, it is not. Not even remotely.  

Quoted:

  Stopped taking you serious right there.


The good thing about this bickering is we're both going to live to see people try this in earnest and the results.


If I'm taking bets, I'll bet on the cold wet fish of reality stinging the optimists more than me.


We'll see.
 

Quoted:


Well... seeing as I have worked on multiple land/air nav solutions that don't use GPS in any way, shape, or form and are currently in production and have been for years, and have been involved with multiple DARPA research efforts and National Research Lab efforts on this topic... I'm gonna stick with GPS is not required. There are many, many ways this can be accomplished without it.

Would you do it without GPS? No. But GPS literally has NOTHING to do with autonomy.

Quoted:


GPS will be secondary and/or tertiary in these systems. I absolutely agree it is not required.

Quoted:


Yep like processing graphics.

Quoted:
Well... seeing as I have worked on multiple land/air nav solutions that don't use GPS in any way, shape, or form and are currently in production and have been for years, and have been involved with multiple DARPA research efforts and National Research Lab efforts on this topic... I'm gonna stick with GPS is not required. There are many, many ways this can be accomplished without it.



Would you do it without GPS? No. But GPS literally has NOTHING to do with autonomy.

Quoted:
That is what I have been talking about the whole time.  Let me clarify a statement I made earlier, "Autonomous LANDNAV is essentially a 2.5DOF problem that actually was "solved" a long time ago, for highly dynamic environments." By that I mean, "a car driving around on streets without GPS. As well as off road - which is actually MUCH more challenging than roads."



Accurate navigation does not require GPS. As stated, it would be tertiary for this situation because it could not be guaranteed. GPS is what is considered the "rough coordinates," if it is even used at all. The GPS info is considered the grossly inaccurate part. Humans managed to find their way around for... well... all of recorded history prior to GPS. Folks have been navigating in cars for a century without GPS. There exist many ways of obtaining positional information without GPS, these methods have been in the "the files are IN the computer!!" digital and even vacuum tube domain for a long time.  Hopefully everyone caught that reference. Hell, I was consulting the Naval Post Graduate school last week on how to refine yet another one of these methods last week. My OT approval came in today to help them on that project again. Yes, this is for driving on real streets... how to all but instantly ascertain your position given ZERO initial positional information.



Originally I thought this was neat news, based on offloading a 9 layer CNN onto a GPU. The amount of calculations required for that is... monumental. However when seeing this was a simplistic (though overfit - aka lol garbage) steering feedback (not even remotely approaching landnav) just using a 3 layer, there is a reason I posted "how the fuck is this even news?"  You can believe it or not, but this has ALREADY existed for a long time and without GPS.





How exactly auto makers roll it out, is the question. I absolutely agree lawyers or .gov can fuck this up with a quickness. Regulations will be the mountain that needs to be climbed, not the tech side.

Quoted:
I think you must have read a different article.  The one you linked talks about an NHTSA "smackdown" but the actual document from the NHTSA was  a major step forward for autonomous cars to recognize that the car, and not an occupant, can be the driver.  It also goes on to talk about how the federal government is set to release the first rules guidance for autonomous cars in six months.  



Liability isn't nearly the issue you make it out to be.  Carmakers already get sued for many deaths and have managed it just fine, despite huge levels of fraud.  Given a grant from the government, I can probably identify 700+ legal issues that should be dealt with based on anyone's daily activity, but that doesn't mean they are significant impediments, or that they need to be addressed at all.



Unions might fight it, but they will lose, just like they have been losing ground for decades.  Insurance companies won't fight it because they will still charge for insurance, increase rates for those who drive themselves, and pay fewer claims.  It's a huge win for them.



You don't have to make any changes in traffic controls or traffic signs.  You are missing the point that it is now feasible to build a computer that can make the same decisions and control the car in the same ways as an average person.  It doesn't need special markings or perfect gps, it can work just like a person does.  There is no magic in a person operating a car well below its limits or navigating to a destination.



It's not just around the corner, most of it is already here.

Quoted:

  HOLY FUCK do you have the ability to talk around me. It's like a Jedi skill.


Please, please, take a deep breath and TRY to understand what I'm saying here. There's a difficulty with autonomy, in the real world, with a car that must navigate by human input (that would be quasi-dumb-shit passengers, btw) on the fly.


Accurate navigation may not require GPS but without mapping/addresses/POI accuracy it's totally fucking meaningless for the daily travel of the 200 million drivers out there what method you're using....you'll just be very accurately pointing to a spot on the globe that is not the place you were wanting to go. And, you might have input a point by which the car is either incapable or very flawed in deciding how to get there, whether the roads are viable or open, or even whether such a point exists at all.


As it stands, the best stuff for compiling all that shit together right now is the open market GPS industry and third-party & open source mapping. And, even that is hilariously shitty and flawed.


Now, one more time, let's talk about driving from one address to another, to possibly a few more, which you may have to put into your car's system on the fly.


Whatever mechanism for getting those coordinates, and then entering them into the car's brains might exist, is fucking ESSENTIAL to SUCCESSFUL autonomy. If you want to play an endless and overly effusive game of semantics over it, or what autonomous means, or how GPS works, dude you're talking around me.


But take careful note, here. The reason that navigation is going to be extremely difficult now and in the foreseeable future is because the same inherent problems with getting accurate info from the real world into the GPS/Map datas that exist will be the same inherent problem as getting them into whatever super cool tech you have going.


For things to work sufficient that autonomous cars don't take a huge sophisticated shit on stuff like this depends on volitional and accurate entry of every single address, description, on every road and building....and a viable and nearly flawless execution of routing and traffic to get there.


That's a huge hurdle.


Let me illustrate this with a small example. The building I work in can't be routed to by the street. Not with GPS solutions, online mapping, even using the city street data. Why? Because it was moved on the property and the entrance now dumps onto a street that is not officially a street yet. There's no data to re-configure this issue, so if you type it in to ANY system and try to get your robo-car to get here, it's going to route you the wrong way down a one-way street, to an entrance that has been removed.


Navigation for a vehicle to accurately deliver you as a driver, needs more than the coordinates of my building (which have also changed, so even if you tried to drop a missle on us, it would miss if your data is out of date, and it likely will be). It needs to know the new place, how the route needs to change, and it can only be as accurate as the input. So, the car will fail, and possibly put itself in a dangerous situation to accomplish this task in today's world.


Are we on the same page yet?

Quoted:

Right.

That would be why Nvidia have a whole range of GPUs that don't actually have a video output and 9 of the top 10 supercomputers are full of them.
http://www.nvidia.co.uk/docs/IO/143900/tesla-k40.png