31–40 of 238 posts
The benefit they point out is 200fps vs. 2000fps, but is that actually that useful? In the post of OpenAO's robot, they said that improving the response times gave no significant gains. Only thing I can imagine would be allowances for wider and deeper neural nets.
Earlier quoted context omitted.
Just wait until they admit that they need LIDAR too... Cha Ching! Various Autopilot accidents strongly imply that whatever the computer “brain” is doing doesn’t matter if it doesn’t have sufficient “eyes” to actually see things like a fire truck, cop car, or bollard.
They seem to have a rather low number of mistakes per billion miles,* so I don't know if more sensors are actually required. I suspect the larger issue is "To Err is Human; To Really Foul Things Up Requires a Computer" https://quoteinvestigator.com/2010/12/07/foul-computer/ The current model makes assumptions that sometimes turn out to be false. So yea it's not going to get to perfection, but level five only needs be…
I am curious on what numbers are you basing this on, the Tesla PR numbers are not valid statistics.
The benefit they point out is 200fps vs. 2000fps, but is that actually that useful? In the post of OpenAO's robot, they said that improving the response times gave no significant gains. Only thing I can imagine would be allowances for wider and deeper neural nets.
36 / 200 frames = 18cm travel distance between each calculation.
So you either can calculate much more each 18cm or calculate shorter distances.
I think a shorter distance is not very useful but calculating more might be.
Earlier quoted context omitted.
They seem to have a rather low number of mistakes per billion miles,* so I don't know if more sensors are actually required. I suspect the larger issue is "To Err is Human; To Really Foul Things Up Requires a Computer" https://quoteinvestigator.com/2010/12/07/foul-computer/ The current model makes assumptions that sometimes turn out to be false. So yea it's not going to get to perfection, but level five only needs be…
That number always comes up, and it always needs to be said that a billion miles with driver assist and a billion miles under autopilot are not the same, and it’s a bogus figure. As to better than a human, it needs to be better than the humans who buy luxury sedans, sober, and awake. Better than a random selection of humans in a random selection of cars is no good. Conflating all vehicle deaths is like conflating all…
See you can try to move the goal post, but as soon as they would be saving lives by taking tired or drunk drivers off the road that's very useful.
I outright snorted at this. Time to market is not why you build a custom chip instead of software. Is this article written by a complete bonehead?
Tesla has it's own hardware platform. Anyone with an ounce of sense will know that was always going to happen because of power consumption, cost and performance. The Nvidia Drive platform is designed to be a quick way to get to market with FuSa. The requirements for fully autonomous modes are estimated by some to be over 10x higher than the top CPU/GPU offerings.
The benefit they point out is 200fps vs. 2000fps, but is that actually that useful? In the post of OpenAO's robot, they said that improving the response times gave no significant gains. Only thing I can imagine would be allowances for wider and deeper neural nets.
More data points can't hurt right? Maybe this is part of their push to move away from third party suppliers?
As a simple example in a different domain, years back I found that the best way to improve Tesseract's OCR accuracy was to ensure that I didn't feed it images at more than 150dpi because it would sometimes misrecognize dust, paper texture, etc. as characters.
Earlier quoted context omitted.
They seem to have a rather low number of mistakes per billion miles,* so I don't know if more sensors are actually required. I suspect the larger issue is "To Err is Human; To Really Foul Things Up Requires a Computer" https://quoteinvestigator.com/2010/12/07/foul-computer/ The current model makes assumptions that sometimes turn out to be false. So yea it's not going to get to perfection, but level five only needs be…
>They seem to have a rather low number of mistakes per billion miles I am curious on what numbers are you basing this on, the Tesla PR numbers are not valid statistics.
I trust insurance companies to have good data, they think it improves safety when you compare rates.
and talking about this now seems like another desperate attempt at generating hype.
Earlier quoted context omitted.
Just wait until they admit that they need LIDAR too... Cha Ching! Various Autopilot accidents strongly imply that whatever the computer “brain” is doing doesn’t matter if it doesn’t have sufficient “eyes” to actually see things like a fire truck, cop car, or bollard.
And yet squishy meatbags manage to sensor fuse a stereoscopic image (plus nascient sound based imaging) into a reliably rendered picture of the world...