Krafcik's quote about how shitty the Tesla cameras are is the only important part:
> "Human vision is so much more capable than the vision of a car equipped with seven 5-megapixel cameras, only one of which is narrow-view, while all the others are wide-view. So you’re basically dispersing those 5 megapixels in a way that makes the actual effective vision more like 20/60 or 20/70. The rest of the cameras in a car like that wouldn’t even pass a DMV vision test"
But we already knew all of this. This is not news. This did not need an article.
The rest is the tired "redundancy" argument, which is also not new:
> "If you have been following the autonomous driving space, you know the debate: Elon Musk believes that since humans drive with eyes (cameras) and a brain (neural nets), cars should be able to do the same. Krafcik, along with the vast majority of the industry, believes that redundancy via LiDAR and radar is non-negotiable for safety."
I really hate this framing.
Redundancy is the least important aspect. Humans just don't drive with "eyes" in the first place. Humans drive with eyes behind lids that squint, inside a head on a neck on a torso on a butt, each part of which has substantial freedom to rotate and move around in space. Your eyes have 6 degrees of freedom plus adaptive light filtering, all of which cyclically respond to the brain's desire to understand the road. Tesla cameras do none of that.
We actually have no idea at all if a vision only approach would work just as well as a lidar system, because nobody has tried a vision only approach that reproduces any meaningful fraction of the human vision system.
On the other hand, a human can only look into one direction, while a Tesla can look into all directions at once. Also human drivers are often distracted, intoxicated, etc. An attentive driver would probably perform better, but I’d rather have all others be slightly worse on the high end of driving skill, while significantly increasing the skill of low end drivers.
> a human can only look into one direction, while a Tesla can look into all directions at once.
Human visual targeting happens extremely quickly relative to the rate of change of objects on the road, so this isn't very important. If it were, attentive driver crashes would happen far more often than they do.
> Also human drivers are often distracted, intoxicated, etc.
This says nothing about the capability of the system. Tesla's have their own non-camera-related failure modes too, like not knowing that traffic lights don't move with the vehicle through space, like changing lanes into opposite-direction lanes, like not having object permanence, like needing to recognize what an object is before deciding to not run it over.
But now also imagine if human drivers were in serious need of glasses but weren't wearing them.
Krafcik's quote about how shitty the Tesla cameras are is the only important part:
> "Human vision is so much more capable than the vision of a car equipped with seven 5-megapixel cameras, only one of which is narrow-view, while all the others are wide-view. So you’re basically dispersing those 5 megapixels in a way that makes the actual effective vision more like 20/60 or 20/70. The rest of the cameras in a car like that wouldn’t even pass a DMV vision test"
But we already knew all of this. This is not news. This did not need an article.
The rest is the tired "redundancy" argument, which is also not new:
> "If you have been following the autonomous driving space, you know the debate: Elon Musk believes that since humans drive with eyes (cameras) and a brain (neural nets), cars should be able to do the same. Krafcik, along with the vast majority of the industry, believes that redundancy via LiDAR and radar is non-negotiable for safety."
I really hate this framing.
Redundancy is the least important aspect. Humans just don't drive with "eyes" in the first place. Humans drive with eyes behind lids that squint, inside a head on a neck on a torso on a butt, each part of which has substantial freedom to rotate and move around in space. Your eyes have 6 degrees of freedom plus adaptive light filtering, all of which cyclically respond to the brain's desire to understand the road. Tesla cameras do none of that.
We actually have no idea at all if a vision only approach would work just as well as a lidar system, because nobody has tried a vision only approach that reproduces any meaningful fraction of the human vision system.
On the other hand, a human can only look into one direction, while a Tesla can look into all directions at once. Also human drivers are often distracted, intoxicated, etc. An attentive driver would probably perform better, but I’d rather have all others be slightly worse on the high end of driving skill, while significantly increasing the skill of low end drivers.
> a human can only look into one direction, while a Tesla can look into all directions at once.
Human visual targeting happens extremely quickly relative to the rate of change of objects on the road, so this isn't very important. If it were, attentive driver crashes would happen far more often than they do.
> Also human drivers are often distracted, intoxicated, etc.
This says nothing about the capability of the system. Tesla's have their own non-camera-related failure modes too, like not knowing that traffic lights don't move with the vehicle through space, like changing lanes into opposite-direction lanes, like not having object permanence, like needing to recognize what an object is before deciding to not run it over.
But now also imagine if human drivers were in serious need of glasses but weren't wearing them.