The video is in the lecture by the CEO of Mobileye that was posted this fall. Really incredible where he showed the deep learning neural networks are capable of detecting the edges of the road even when lane markings are covered in snow.
-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
Full autonomy in 2-3 years
- Thread starter paulj
- Start date
-
- Tags
- Model S User Interface
The video is in the lecture by the CEO of Mobileye that was posted this fall. Really incredible where he showed the deep learning neural networks are capable of detecting the edges of the road even when lane markings are covered in snow.
Skotty
2014 S P85 | 2023 F-150L
Yay! Someone who agrees with me. Keep pushing the boundaries, sure, but this is going to be a lot harder than most people think. It's hard because our roadways are not designed to be navigated by machines. To drive on today's roads, you have to be able to make sound decisions off of incomplete information, potentially dealing with unknown variables you don't expect, make judgement calls, and it also helps to care about the outcome of your decisions. That is not something a computer is good at.
Thinking of how you would do it safely...
A brand-neutral central network operating off cell towers collects information from cars in real time, and releases information in real time. Cars build a 3D wireframe constantly about their surroundings, all 360°. It transmits what it "sees" as a wireframe, and fills in areas it can't see by polling the network. The car has a large SSD in it to buffer it's wireframe concerning fixed objects. The cars are sold with regional information pre-installed.
There are at least 3 classes of wireframes. Fixed objects that are permanent. Fixed objects that are temporary (like construction or snow banks). Then movable objects.
Your car's video cameras create a 3D wireframe of Class 3 objects then ties into the Class 1 and Class 2 wireframes to determine their location.
Sort of like a multiplayer 3D video game.
- - - Updated - - -
Keep in mind a car, truck, person, dog, is simply 8 XYZ double precision floats defining a polygon of it's envelope dimensions. These can be calculated quickly, and use very little bandwidth.
A brand-neutral central network operating off cell towers collects information from cars in real time, and releases information in real time. Cars build a 3D wireframe constantly about their surroundings, all 360°. It transmits what it "sees" as a wireframe, and fills in areas it can't see by polling the network. The car has a large SSD in it to buffer it's wireframe concerning fixed objects. The cars are sold with regional information pre-installed.
There are at least 3 classes of wireframes. Fixed objects that are permanent. Fixed objects that are temporary (like construction or snow banks). Then movable objects.
Your car's video cameras create a 3D wireframe of Class 3 objects then ties into the Class 1 and Class 2 wireframes to determine their location.
Sort of like a multiplayer 3D video game.
- - - Updated - - -
Keep in mind a car, truck, person, dog, is simply 8 XYZ double precision floats defining a polygon of it's envelope dimensions. These can be calculated quickly, and use very little bandwidth.
Mark Z
Active Member
When I think of autonomous vehicles in two years, IMHO it would be with hands on the steering wheel to monitor the computer actions.
I would expect the vehicle to travel the distance without user action, but would require user intervention if necessary.
In today's Model X, I am amazed at how much Autopilot does. Turns and stops will take time to perfect. I don't mind doing that manually for a few more years!
I would expect the vehicle to travel the distance without user action, but would require user intervention if necessary.
In today's Model X, I am amazed at how much Autopilot does. Turns and stops will take time to perfect. I don't mind doing that manually for a few more years!
EcoHeliGuy
Member
How come Tesla give as much as it has in 2 updates in a couple months and everybody thinks it's still years away? To my knowledge no one else is selling an automatic cruise control anywhere near teslas capabilities yet.
They could have full autonomy on the highway *now* with an upgraded sensor suite. Driving around the city is orders of magnitude harder than highway driving.
I'm guessing it's because everyone has a different expectation of what full autonomy means. There is a thread devoted to the increased nag in 7.1. Until the nag is reduced, I don't think it's as close as some people think.
Actually there are quiet a few - like Mercedes, VW, Infinity... it is rather old tech from 2014.
Infiniti Q50 Active Lane Control - Selfdriving Car - YouTube
Romanian driver shows off tricking autopilot mode in car - YouTube
Volkswagen Phaeton W12 (Team Review) - Fifth Gear - YouTube
EcoHeliGuy
Member
Those systems don't change lanes, accelerate for passing, decel for off ramps, come too a full stop behind traffic and then proceed with the the flow of traffic, from my understanding. All features of Tesla's TACC
Recently there was an article in the local paper about traffic/mass transit planning. Part of the article was discussing the potentially large impact self-driving cars could have on traffic, and how the best near term strategy would be to add more HOV lanes on all of the major highways, and within the next 10 years if those lanes are restricted to self-driving cars, how much additional traffic they'd be able to handle (assuming self-driving cars would be able to safely maintain higher speeds, resulting in increased capacity on the HOV lanes).
As mentioned above, while the technology may be available to implement fully self-driving cars within a few years, it could take considerably longer before the manufacturers (including Tesla) will enable that feature, because of the potential liability issues. Today, unless there's a flaw in the car, if there's an accident, the driver is responsible. With a self-driving car, that responsibility could shift to the manufacturers. And when the manufacturers are concerned about liability, they impose restrictions - like how most nav systems (other than Tesla) restrict functionality while the cars are in motion (because they don't want to be liable for an accident).
As mentioned above, while the technology may be available to implement fully self-driving cars within a few years, it could take considerably longer before the manufacturers (including Tesla) will enable that feature, because of the potential liability issues. Today, unless there's a flaw in the car, if there's an accident, the driver is responsible. With a self-driving car, that responsibility could shift to the manufacturers. And when the manufacturers are concerned about liability, they impose restrictions - like how most nav systems (other than Tesla) restrict functionality while the cars are in motion (because they don't want to be liable for an accident).
No, Tesla really can't have full autonomy now, even with an updated sensor suite (redundancy on every step, bi/tri-focal cameras, side cameras, rear radar, etc.). They don't have the software for it yet, and even though Autopilot works great, it had many limitations.
Now if you're suggesting a crippled Level 3 that only works in perfect weather, with perfect lane markings, etc. then yes, Tesla can get there very quickly if they update the hardware.
- - - Updated - - -
Tesla doesn't change lanes on it's own. The decel for off ramps is somewhat annoying, as it can start to decel randomly (not necessarily at an offramp).
Otherwise I think the Inifnit Q50, Mercedes and Acura systems are pretty close to Tesla's. They all have nags now.
- - - Updated - - -
Once self driving cars are a reality, full autonomy -- say goodbye to mass transit (sure it wont happen right away, and the mass transit that exists may stay in place, but no need to further bulid it out, especially when the price of the technology goes down).
No need for buses or trains. Hail a driverless uber and get to where you need to go. Uber can even have an unlimited plan, just like the subway has an unlimited ride plan -- get to/from work, door to door, in a city, without having a car or using the trains.
It will be a while before there are enough self-driving cars on the road to have a large impact on traffic.
But when that critical mass is achieved and the shift begins to restrict highway access to self-driving cars, the current highway/HOV infrastructure should be able to handle more cars - driving at higher speeds - and more safely - than human drivers today.
Self-driving taxis would be much more cost effective than most of the mass transit solutions today - that often have large vehicles only partially filled with passengers. Plus self-driving taxis can drive directly to the destination, eliminating the need for transfers with a spoke-and-hub mass transit system.
Plus, the auto industry will likely see fewer cars purchased for private use - and a higher percentage of cars purchased for shared use.
But we're years away - and the recent article I saw was recommending that mass transit planning should start shifting now to anticipate self-driving cars.
But when that critical mass is achieved and the shift begins to restrict highway access to self-driving cars, the current highway/HOV infrastructure should be able to handle more cars - driving at higher speeds - and more safely - than human drivers today.
Self-driving taxis would be much more cost effective than most of the mass transit solutions today - that often have large vehicles only partially filled with passengers. Plus self-driving taxis can drive directly to the destination, eliminating the need for transfers with a spoke-and-hub mass transit system.
Plus, the auto industry will likely see fewer cars purchased for private use - and a higher percentage of cars purchased for shared use.
But we're years away - and the recent article I saw was recommending that mass transit planning should start shifting now to anticipate self-driving cars.
EcoHeliGuy
Member
Once we have cars that don't crash anymore. A lot of the bloated weight and mass from recent years can be lost as well. This will be further down the road after some proving time.
the Inifnit Q50, Mercedes and Acura systems are pretty close to Tesla's. They all have nags now.
[/QUOTE]
The current crop of competitors' systems are nowhere near close to Tesla's in capability - they have error rates over 100% higher - look up the fascinating comparison test results in Car and Driver last month. Having said that, the 2017 Mercedes E-Class is taking the game to the next level and may well beat Tesla's current system in real world performance tests - it will be fascinating to see whether MBZ's superior hardware suite will out-perform Tesla's inferior current hardware. Or perhaps Tesla's fleet learning will have really improved to the point that even with only one camera and one radar gun Tesla's will make fewer errors than MBZ's.
[/QUOTE]
The current crop of competitors' systems are nowhere near close to Tesla's in capability - they have error rates over 100% higher - look up the fascinating comparison test results in Car and Driver last month. Having said that, the 2017 Mercedes E-Class is taking the game to the next level and may well beat Tesla's current system in real world performance tests - it will be fascinating to see whether MBZ's superior hardware suite will out-perform Tesla's inferior current hardware. Or perhaps Tesla's fleet learning will have really improved to the point that even with only one camera and one radar gun Tesla's will make fewer errors than MBZ's.
Then I guess we'll have to agree to disagree. I've seen the car and driver article.
Tesla is in the lead, but the competitors aren't far off (hence my pretty close statement) wrt autopilot. It's no longer ground breaking/bleeding edge stuff.
We'll see if Tesla will be the first to come out with autonomous driving or not.
Tesla is in the lead, but the competitors aren't far off (hence my pretty close statement) wrt autopilot. It's no longer ground breaking/bleeding edge stuff.
We'll see if Tesla will be the first to come out with autonomous driving or not.
ratsbew
Active Member
I think the next step will be highway autonomy where you can input a destination in the nav system and the car will make its own exits and on ramps. I think that surface streets will be more like 10 years off (although Google does it today, so maybe not that far...).
The self-driving bubble shaped Google cars are capped at 25 mph (mentioned at FAQ – Google Self-Driving Car Project). They've had self-driving Priuses and Lexus RX SUVs for many years now and they are NOT capped at 25 mph.
For example, you can see references to highways and 63 mph at https://static.googleusercontent.com/media/www.google.com/en//selfdrivingcar/files/reports/report-0515.pdf. Google: Self-driving cars are mastering city streets - CNN.com and many others from around that era from 2014 state "Since 2011, when self-driving vehicles became street-legal in Nevada, Google has logged nearly 700,000 miles with the cars, mostly on highways..."
Similar threads
