The problem with that approach is that self-driving companies have their models trained for their specific hardware, which will differ a lot and make the data i

The problem with that approach is that self-driving companies have their models trained for their specific hardware, which will differ a lot and make the data incompatible. The only way around that would be to enforce a standard platform, similar to what happens in racing, leaving the differences between competitors to the software that drives it.