-
Want to remove ads? Register an account and login to see fewer ads, and become a Supporting Member to remove almost all ads.
To be frank, does it matter? I'll assume your asking about the sensors due to your interest in autopilot. In that regard, I would bet the most profound improvements are to come with software updates and not new sensors. I remember an exercise in a robotics class I took where the only sensor you could use was a small ultrasonic rangefinder and the goal was to come up with the most valuable use of that one sensor. One team ended up being able to map a small room, another could judge the hardness of an object it was aimed at, another could make a rough 2D image of an object if it was large enough, and all of this was done with a cheapo sensor because the main work was done in software.
vjason
Member
Until someone comes up with a way to very accurately measure road position in a snowstorm, especially when no other cars are around you, all sensors used with autonomous driving today are somewhat limited.
Ford is developing very detailed maps to try and do this (The Clever Way Ford’s Self-Driving Cars Navigate in Snow) but the US is rather large (as are many countries around the world) so I would't expect this to work for anything but highways and large cities (for now).
Perhaps in the future they will embed sensors in roadways and the cars can read those. Again, expect that on highways and big cities first.
It is nice to want the latest and greatest, but I don't think the tech is quite there yet to enable door to door autonomous driving in all possible conditions.
FWIW, lots of data is needed either way so obviously Tesla is helping out there. That being said, we are a ways off from true autonomous for more reasons than just snowstorms (talking legislation , insurance, liability, etc, etc).
Ford is developing very detailed maps to try and do this (The Clever Way Ford’s Self-Driving Cars Navigate in Snow) but the US is rather large (as are many countries around the world) so I would't expect this to work for anything but highways and large cities (for now).
Perhaps in the future they will embed sensors in roadways and the cars can read those. Again, expect that on highways and big cities first.
It is nice to want the latest and greatest, but I don't think the tech is quite there yet to enable door to door autonomous driving in all possible conditions.
FWIW, lots of data is needed either way so obviously Tesla is helping out there. That being said, we are a ways off from true autonomous for more reasons than just snowstorms (talking legislation , insurance, liability, etc, etc).
favo
P3D+ owner
Detailed maps and GPS-based databases can help, but I think better vision algorithms are the key. Ultimately, that is what people use, so it should be good enough (easier said than done, of course). Mobileye, which so far is Tesla's supplier of vision processing chips (EyeQ), has been working on the issue of poor visibility and/or poor lane markings. See starting around 22:32 in this video. Their new EyeQ4 chip is supposed to be much better in this regard, with better hardware, software, and more sensor capabilities.
More sensors with more redundancy are key to full autonomy. You can't have the car give up driving, just because a bug hit the front camera lens.
More sensors with more redundancy are key to full autonomy. You can't have the car give up driving, just because a bug hit the front camera lens.
Canuck
Well-Known Member
Canuck
Well-Known Member
Mobileye has stated that their next architecture will need three forward cameras, and then ideally have side ones as well. So, yes, I would fully expect the Model 3 to have an enhanced sensor suite, as will the Model S and X in two years.
As far as precise maps go, Mobileye (and presumably Tesla) are working on those too. It isn't what you think of when you think of a "map". The problem with GPS is that it isn't accurate enough to let you know where you are. It is probably good to a few meters or yards, but that's way, way too inaccurate for positioning within a snowy lane. So what they will/are doing is building a database of continually updated maps of road signs. You can't drive 100 feet in the US without hitting some sort of road sign, be it a highway curve ahead sign, a speed limit sign, a wildlife crossing sign, etc. With the cameras, the car positions itself in space relative to these signs that it has a database of. This technique, along with dead reckoning between the signs, can localize a car within inches.
As far as precise maps go, Mobileye (and presumably Tesla) are working on those too. It isn't what you think of when you think of a "map". The problem with GPS is that it isn't accurate enough to let you know where you are. It is probably good to a few meters or yards, but that's way, way too inaccurate for positioning within a snowy lane. So what they will/are doing is building a database of continually updated maps of road signs. You can't drive 100 feet in the US without hitting some sort of road sign, be it a highway curve ahead sign, a speed limit sign, a wildlife crossing sign, etc. With the cameras, the car positions itself in space relative to these signs that it has a database of. This technique, along with dead reckoning between the signs, can localize a car within inches.
Canuck
Well-Known Member
That's not the issue. Of course it can be done. Driving through intersections can be done with the current AP 1.0 hardware. But in my opinion you will never see it happen because the liability exposure is far too great given the limitations of the current hardware. We will only see it happen when Tesla finally gets around to checking these last three boxes:
Here you go - as I predicted, three forward cameras, exactly in the configuration of Mobileye's next gen architecture.
New Tesla Model S Has 2nd Triple Cam For Autopilot & Pedestrian Noise Unit (CleanTechnica Exclusive)
New Tesla Model S Has 2nd Triple Cam For Autopilot & Pedestrian Noise Unit (CleanTechnica Exclusive)
Similar threads
