Well, given now that NHTSA and IIHS have dropped their top ratings for Tesla until the new Tesla Vision system can be tested, all the pressure is on Tesla now. This means actually delivering a system, which I don't believe any of Elon's tweets. Second, they have to stop hitting large objects on the highway. Failed this test with radar, and supposed to be better with all camera? I'm sorry, but it is getting very difficult to believe this is going to end well, and for the price they are asking, we STILL don't know how this system will function for basic tasks, especially in bad weather, when sun is blinding the cameras, etc. I can't believe I'm saying this, but Ford, VW, and GM really have a chance to catch up or even leap-frog if this goes sideways for Tesla.
Training a computer vision model to perform well in a range of scenarios is challenging. Training it to cover all edge cases is close to impossible. Too many variables, and I'm not even accounting for software bugs/failures. Blinding sun, reflecting objects such as large white box trucks, snow, fog, etc. in my opinion will render the trained computer vision model useless, or at least so inaccurate that it is useless. There is no new AI/ML magic here. Tesla is training computer vision models, which create bounding boxes around objects in which the model has high confidence it detects, and then software makes choices based on what these CV models "see", such as what it believes the object is, and what action to take. You can't train a model for every scenario, even with billions of miles of driving data. Without radar as a backup sensor, which is already secondary to LiDAR, the choice to go all-in feels to me like "we're committed to cameras come hell or high water." No other vendor in the world seems to be going this direction, and even worse, legislation, which is out of Tesla's control, could squash their best-laid plans, no matter how much Elon sends angry tweets. I would love to see Tesla succeed here, but it is hard not to be concerned. I'm also wondering if my June 2020 build will abandon the use of the existing radar systems in the next update. If that is the case, and the system isn't working out well, watch the trolls react, and I'd seriously consider moving to another brand at that point, and hope that my resale value doesn't drop to the floor. I love my MY, but I need the safety systems to work well, and work predictably.