Can you elaborate please?Fun fact: They are developing the self driving stuff completely wrong.
Can you elaborate please?Fun fact: They are developing the self driving stuff completely wrong.
As long as your hardware can support her
As long as your hardware can support her
They are using a neural network to try and classify things then use software to try and predict where they are going instead of using an end to end model that handles everything in the network. Think about it, if you are driving by a bike rack with a hundred bikes that is going to have to try and classify each one and model the physics behind each one. That takes a lot of power. They are also trying to create "HD maps" for their cars to follow, which is just silly and will always be out of date. One of the times they hit someone was coming out of a bike rack, they programmed the car to ignore areas with bike racks to save all that processing power they are wasting.Can you elaborate please?
They are using a neural network to try and classify things then use software to try and predict where they are going instead of using an end to end model that handles everything in the network. Think about it, if you are driving by a bike rack with a hundred bikes that is going to have to try and classify each one and model the physics behind each one. That takes a lot of power. They are also trying to create "HD maps" for their cars to follow, which is just silly and will always be out of date. One of the times they hit someone was coming out of a bike rack, they programmed the car to ignore areas with bike racks to save all that processing power they are wasting.
Comma is using an end to end model, and that is why they are better than everyone else right now. This includes Tesla, they don't have a couple of features that Tesla has but the important one is coming soon and should be here by the end of the year.
Saying end to end model is pretty vague though, I'm not following the difference Comma has. If they don't do all that processing, the physics of all scanned vehicles, HD maps and whatnot like you listed, then what exactly are they doing?
The neural network stuff is all vague, it's hard to know what they are actually doing. The gist of it is they train a model based off of the current and previous frame as well as where the driver is driving based on recorded inputs of the driver from the CAN bus. The model then outputs where the car should go based purely on the model. The model figures out what is important based on all of that data that is being used to train it.
Another advantage Comma had over Tesla was driver monitoring. They use IR LEDs and a camera that can accept IR for it. Tesla has since started to do the same thing however.
So like, they have a real driver drive in the locations they think people will drive in and record that driver's movements so that it will go similarly when someone drives in that same area? I reckon though there's still that processing you mentioned before for collision detection and whatnot, a bike rack can be anywhere as well, though I guess it's less processing if it's only for select things. Assuming I understood.
Though honestly I can't tell if it's still better. Even if it's more intensive, all that processing should in theory be more dynamic and be able to respond to situations intelligently which sounds safer. As this thing is meant to be widespread moreso in the future, processing power should be able to catch up as well to give more headroom.
It makes sense to make it work right, then work on making it more efficient, people's well being is at risk here.
I just wanna game related memes! IA or not!!!