Another option is to stop trying to teach cars to see (arbitrary object identification with computer vision has been a work-in-progress for 60 years, and has yet to be genericized), and build detection & guidance systems into the road, making the localization trivial and without incurring new in-vehicle equipment expenses.
This provides multiple benefits:
- Reducing the consumer burden of everyone purchasing their own computer vision systems, which add significant cost to vehicles
- Amortizing the expense of vehicle localization across all users, instead of individually as above
- Reducing the total systemic cost of obtaining autonomy by enabling the sharing of systems
- Improving roads, which can we please admit, autonomous cards need anyway
Independent-autonomy “solutions” don’t do squat to address the state of the network (of roads) that the cars depend on, and it is incredibly frustrating that people refuse to admit that autonomous cars inherently require a network of roads, and a network for communication.
Implementing a network to deliver these kinds of services to inexpensive network-dependent autonomous vehicles could be delivered in a shorter timeframe than delivering an independent autonomous vehicle at today’s average vehicle pricing of $25000.
Do we want good roads that everyone can use, which support inexpensive autonomous vehicles?
Or do we want wildly expensive autonomous vehicles that only a few people can afford, operating on our beleaguered infrastructure?
For me, the choice is obvious, if we can just get over the reluctance of technologists to accept that roads are necessary and inevitable.