Thank you for the comment.

It's not all of the real-world. It's over 99% do to the massive volume scenarios to run and rerun. Shadow driving is necessary for gathering data and intention testing. As well as to inform and validate the simualtion. But we need to do a lot less of it to be able to finish. With regard to public safety driving virtually all of that should go.

Your "never dream up" point is exagerated and largely irrelvant. Exagerated because there are tons of data sources including the shadow driving we can do to make enough primary scenarios to then derive enough varations to get to a provable sigma point where AVs are safe enough. With regard to the irrelavant part. If I can't spend a literal eternity or anywhere close to $300B per company I can't avail muyself of but a fraction of what you say needs to be found.

It comes down to infinty of time and cost (basilacally,. to many injured or dead human Guinea pigs and a perfect data set I can't get a fraction of vs getting just enough of that to then extrapoloate enough derivitives to mitgate the time, cost and dead human Guinea pigs issue.

The DoD/aerospace sim tech part is to be able to build an accurate enogh digital twin to replace enough of the real world to get to that good enough point. Gaming tech has far too many architectural and model fidelity issues to do that. They look great but the physics and real-time ops come up way to short.