I've been trying to visualize the exponential growth of FSD's abilities that Elon continually alludes to, and I think I finally figured out how to think about this concept.
I believe the biggest thing people are missing with Tesla's FSD tech is that it doesn't require lane markings to make its decisions. Think about how you would typically make a decision when you're driving down a street with no markings (imagine a small town single lane road with no yellow or white line down the middle of the road). You are using spatial awareness to navigate down this road by using the relative distance of your position to the edge of the road and the approximate middle. This is what the neural net is being built to do.
Other car manufacturers are building self-driving tech that's heavily reliant on lane markings, signage, etc. Remember that lane markings exist because humans are not perfect and we need to constantly cater to the lowest common denominator instead of trusting that folks will figure out how to navigate thru miles of asphalt. Think of a new driver that's driving around a town with awesome lane markings and all of a sudden they transition into a small town road that's just gravel. The "oh *sugar*" factor increases by a factor of 10. The difference between Tesla's self-driving tech and other manufacturers is going to be directly proportional to the relative strength of a very good driver that knows how to navigate roads using spatial awareness vs. one that's heavily reliant on road markings.
All the training and investment of time in the tech is to teach the system to be aware of where it can and can't drive, as Elon mentioned more than once on Autonomy Day, regardless of road markings. Again, think about how you would typically drive in the snow - you rely on previous car's tire tracks and nearby landmarks to guide you. The lane markings are no longer there to use, but somehow our brain is able to say "alright, if we stay in the relative neighborhood of these tire tracks when there's a bunch of white stuff on the road and not get too close to that snow bank we should be OK".
I believe the amount of skepticism out in the media and internet boils down to folks not being able to wrap their minds around this concept. The machine is literally learning how humans use spatial information, not what the markings are telling humans to do. This is why Tesla's cars on the road with the hardware that's collecting data is such a massive advantage and growing. It's learning how humans work, and honing itself to be the best driver that exists using all of our strengths.
This period in Tesla reminds me very much of when the company was transitioning from the Roadster to the Model S. The amount of skepticism was through the roof, yet a couple years later Tesla delivered a show stopper. I think Tesla is set to do this again, but at a scale that will impact the entire globe all at once. Youtube, Twitter, Instagram, and Facebook will be a collection of millions of self-driving videos for all to see on day one.
I believe the biggest thing people are missing with Tesla's FSD tech is that it doesn't require lane markings to make its decisions. Think about how you would typically make a decision when you're driving down a street with no markings (imagine a small town single lane road with no yellow or white line down the middle of the road). You are using spatial awareness to navigate down this road by using the relative distance of your position to the edge of the road and the approximate middle. This is what the neural net is being built to do.
Other car manufacturers are building self-driving tech that's heavily reliant on lane markings, signage, etc. Remember that lane markings exist because humans are not perfect and we need to constantly cater to the lowest common denominator instead of trusting that folks will figure out how to navigate thru miles of asphalt. Think of a new driver that's driving around a town with awesome lane markings and all of a sudden they transition into a small town road that's just gravel. The "oh *sugar*" factor increases by a factor of 10. The difference between Tesla's self-driving tech and other manufacturers is going to be directly proportional to the relative strength of a very good driver that knows how to navigate roads using spatial awareness vs. one that's heavily reliant on road markings.
All the training and investment of time in the tech is to teach the system to be aware of where it can and can't drive, as Elon mentioned more than once on Autonomy Day, regardless of road markings. Again, think about how you would typically drive in the snow - you rely on previous car's tire tracks and nearby landmarks to guide you. The lane markings are no longer there to use, but somehow our brain is able to say "alright, if we stay in the relative neighborhood of these tire tracks when there's a bunch of white stuff on the road and not get too close to that snow bank we should be OK".
I believe the amount of skepticism out in the media and internet boils down to folks not being able to wrap their minds around this concept. The machine is literally learning how humans use spatial information, not what the markings are telling humans to do. This is why Tesla's cars on the road with the hardware that's collecting data is such a massive advantage and growing. It's learning how humans work, and honing itself to be the best driver that exists using all of our strengths.
This period in Tesla reminds me very much of when the company was transitioning from the Roadster to the Model S. The amount of skepticism was through the roof, yet a couple years later Tesla delivered a show stopper. I think Tesla is set to do this again, but at a scale that will impact the entire globe all at once. Youtube, Twitter, Instagram, and Facebook will be a collection of millions of self-driving videos for all to see on day one.
