Will current Teslas never get linked to navigation because of limited rear vision?

[calisnow]

Since autopilot has very limited rear "vision" what are the odds that current autopilot Teslas will never get linked to GPS navigation instructions because they aren't capable of safely changing lanes (such as to take an exit specified by the navigation system on a route) without a human to check first?

And if that is the case, what do you think the odds are that we are in for a sensor suite upgrade in a year or less?

I almost wonder if the lack of the obviously needed rear vision is an almost Apple-like marketing strategy of holding back key features until the next "release."

Sorry if this has already been beat to death - but I NEVER lease cars - but I'm almost ready to lease a Tesla instead of purchasing because they are changing so fast.

 

[robertvg]

You are right. You will need the next version of MobileEye that can handle 8 camera's to create a realtime 360 view around your car. And even then I'm not sure if I agree with Elon that you can do without something like lidar.
But I would not wait for a sensor suite upgrade, it will be almost impossible to build in something like that.
Better start saving for a new car :smile:

 

[mgboyes]

Fully autonomous driving on the public highway is a decade away at least, in particular for regulatory reasons. And the current Model S hardware is probably 2 or 3 generations behind what will be needed to accomplish that and satisfy regulators that it really is more capable than a human.

Even if there is a sensor upgrade in the next year (the Model X renders show what appears to be stereo front cameras, even though the founders' cars didn't have them) then I doubt it will add any significantly more amazing features than we have today. You are not going to be able to stop paying attention to the road for at least 10 years.

 

[sillydriver]

I had assumed the current cars would not be upgradable but now I wonder. I believe the current MobilEye EyeQ3 processor is in the camera housing on the windshield, which can be replaced. I saw a very interesting (though 9 month old) talk by the MobilEye CEO on YouTube. He said the new EyeQ4 works best with three cameras pointing straight forward: 180 degree fisheye, 50 degree view and 25 degree view, which I assume can all be in the replacement windshield camera housing. As for the backwards 180 degrees, why not the existing fisheye rear-view camera? The challenge is whether the windshield camera housing is connected to a data bus that could carry the video from that rear camera to the housing on the windshield. If Tesla had thought ahead, it might. I hope my theory is correct!

 

[Max*]

Fully autonomous driving on the public highway is a decade away at least, in particular for regulatory reasons. And the current Model S hardware is probably 2 or 3 generations behind what will be needed to accomplish that and satisfy regulators that it really is more capable than a human.

Even if there is a sensor upgrade in the next year (the Model X renders show what appears to be stereo front cameras, even though the founders' cars didn't have them) then I doubt it will add any significantly more amazing features than we have today. You are not going to be able to stop paying attention to the road for at least 10 years.

Isn't the MobilEye Q4 chip capable of full autonomous highway driving (Level 3), and didn't MobilEye come out and say they think Level 3 is 3-5 years away?

 

[sillydriver]

Isn't the MobilEye Q4 chip capable of full autonomous highway driving (Level 3), and didn't MobilEye come out and say they think Level 3 is 3-5 years away?

I hope you have seen the video because I think it contains a lot of nuggets: search YouTube for MobilEye, it's about half an hour long, posted Jan 15. Looking at my notes, he said 2015 - 16 was 'highway auto-pilot', 2017 - 18 was 'city traffic', and 'true autonomous' driving was 'later' (yes, I got the sense that 3 - 5 is right) and would benefit from more detailed maps and car-car communications. EyeQ4, and even the current EyeQ3, doesn't use all its capacity running the current algorithms. EyeQ4 performance is about 10x EyeQ3. He said the two basic strategies to implement automated driving are at one end of the spectrum 'store & align' with maps, which is the Google strategy, and at the other end of the spectrum 'sense & understand' which is the MobilEye strategy, which is independent of maps. It sounds like Tesla's strategy is in the middle of the spectrum, using both. I though it interesting that his 'more detailed maps' and 'car-car' comment implied that the solution to the autonomy problem would require both strategies.

 

[ItsNotAboutTheMoney]

Yes, it's been beaten to death.

The current hardware is not capable of full autonomy. However, it's capable enough to make driving less physical and mentally demanding and should improve over time.

I see Tesla taking the approach they did with Autopilot: do a lot of work, figure out what hardware is needed for the next step and then add the hardware once they can deliver some benefit, with promise of future software updates.

The one significant potential hardware change that I think could come early would be side cameras (with emergency back-up mirrors), because they could raise efficiency signifcantly. But that depends on the NHTSA allowing them to use them.

 

[LetsGoFast]

Since autopilot has very limited rear "vision" what are the odds that current autopilot Teslas will never get linked to GPS navigation instructions because they aren't capable of safely changing lanes (such as to take an exit specified by the navigation system on a route) without a human to check first?

And if that is the case, what do you think the odds are that we are in for a sensor suite upgrade in a year or less?

I'm not convinced that the lack of a rear camera has any particular relevance to the inability to take an exit ramp. Changing from the center lane to the left lane is a much higher risk than changing from the right lane into the exit lane. So, I'd suspect that at a minimum we will be able to manuever into an exit lane by using the turn signal in the relatively near future. I'm not counting on the car automatically taking the exit with no intervention at all anytime soon, because there would be very little point to that if the car was not yet trustworthy in city driving.

I do expect to see a sensor suite upgrade in the relatively near future (less than one year from today). This will undoubtedly include more front-facing cameras and probably an additional rear camera (with some change of simply replacing the existing one with a better camera). If they follow true to the past upgrades, it won't be very practical to upgrade existing cars (it will likely require new wiring harnesses).

 

[pilotSteve]

Yes, it's been beaten to death.

The current hardware is not capable of full autonomy. However, it's capable enough to make driving less physical and mentally demanding and should improve over time.

I see Tesla taking the approach they did with Autopilot: do a lot of work, figure out what hardware is needed for the next step and then add the hardware once they can deliver some benefit, with promise of future software updates.

The one significant potential hardware change that I think could come early would be side cameras (with emergency back-up mirrors), because they could raise efficiency signifcantly. But that depends on the NHTSA allowing them to use them.

Haha you sound like one of "classic" owners disappointed we can't upgrade to the current autopilot! I'm resigned to enjoy what I have and buy the latest every four or five years.