I'd say that Kalman filters are perfect for this-- they optimally blend estimates from multiple sources, including a model of the system dynamics (where you thi

I'd say that Kalman filters are perfect for this-- they optimally blend estimates from multiple sources, including a model of the system dynamics (where you think you're going to be next based on where you were before, i.e. "I'm not going to suddenly jump 100 miles away"), and all the various sensor inputs ("GPS says I'm here, cameras say I'm here, wheel distance says I'm here"). On top of that, Kalman innately tracks the uncertainty of its combined estimate. So you can simply look at the Kalman uncertainty covariance and decide if it is too much for the speed you are going. I really wonder if Tesla is doing that...