Self-driving is not going to be solved by throwing money at it. If it was, Google would have "solved" it already and Apple wouldn't have abandoned their effort.
The obvious shortage is engineering talent. This is more debatable (many people, like Elon Musk, don't believe this to be the case), but the field also needs one or more research breakthroughs, especially in the realm of planning and reasoning. Most of the machine learning in self-driving cars goes towards perception / detection / mapping. The actual decision-making is still heuristics and hard-coded rules.
Network and data infrastructure is also a challenge for self driving cars and has a ways to go. The data produced by a single self driving car can be pretty staggering (I've heard estimates as high as 5 TB/minute). Now think about transferring that much data around a fleet of cars over current LTE wireless data connections.
Even if someone gets it working in a small area, it will take years of investment in network infra to make it truly ubiquitious.
Barring some major technical breakthroughs, I just don't see the current tech that these companies are testing being scalable at the same level as GPS, cellular data, etc.
> The actual decision-making is still heuristics and hard-coded rules.
The planning system always be hard coded rules. Perception, fine, throw the latest CNN at it. Planning and control, better stick to the DMV/NHTSA traffic book.
Self-driving cars are going to produce >$100B/year in value (probably at least $1T/year worldwide). They are going to cost a lot more than $5B to develop. Google has >$80B of cash on hand and Apple has >$230B. Those companies would spend $5B to develop self-driving cars in a heartbeat if they could, but they can't.
The obvious shortage is engineering talent. This is more debatable (many people, like Elon Musk, don't believe this to be the case), but the field also needs one or more research breakthroughs, especially in the realm of planning and reasoning. Most of the machine learning in self-driving cars goes towards perception / detection / mapping. The actual decision-making is still heuristics and hard-coded rules.