Rocketry and self-driving are completely different domains of difficulty. Rocketry is a solved problem - the basic math works, we've done it before, the challenge now is building upon that knowledge, refining it and increasing efficiency, but the basics have been figured out and proven since we've launched the first satellites in orbit. SpaceX is successfully doing this.
Unsupervised self-driving in an unconstrained environment is a completely different game. The reason humans can drive is that they have millions of years of evolution in both low-level image processing, reasoning, etc, diverse knowledge of the world that an individual obtains during the course of their life which can be blended in various ways to make driving-related decisions, as well as a sense of self-preservation that generally forces them to err on the side of caution. Short of a major AGI breakthrough, safe self-driving on existing road infrastructure is impossible.
> Rocketry is a solved problem - the basic math works, we've done it before, the challenge now is building upon that knowledge
This is not true. The only thing that enabled the reusability of SpaceX rockets was the convexification of the soft landing problem, a novelty by Lars Blackmore.
Sure, you could argue the the model predictive control scaffolding was pre-existing, but so was linear algebra for self driving.
Unsupervised self-driving in an unconstrained environment is a completely different game. The reason humans can drive is that they have millions of years of evolution in both low-level image processing, reasoning, etc, diverse knowledge of the world that an individual obtains during the course of their life which can be blended in various ways to make driving-related decisions, as well as a sense of self-preservation that generally forces them to err on the side of caution. Short of a major AGI breakthrough, safe self-driving on existing road infrastructure is impossible.