Self Driving Cars: Robotics R&D program mistaken for a business opportunity

The technical, social, and political risks to commercially viable self-driving cars are underestimated. At the same time, the economic benefits are overestimated.

I've commuted SF to MTV in the past (1-2 hours each way), and can say that while taking a well-appointed bus is certainly better than driving, it's still 70% of the pain. The driving part isn't the reason we don't enjoy being in moving vehicles for extended periods of time. Space / weight constraints are the real reason – you can't have a kitchen / TV / comfortable chair / friends / yoga on limited road infrastructure. This has nothing to do with whether the driver is meat or silicon based. Basic economics says SDCs will increase commute times right up until they've offset reduced pain / cost from not driving, which may not be much.

More problematic, the corresponding increase in road traffic and commute times will affect human driven cars too. They will not be happy about it, and it will easily (and accurately) be framed as a rich elite vs working middle class issue. It's not hard to imagine how that will play out politically.

Self-driving cars only get exciting when we can effect dedicated rules / infrastructure for them that increases throughput on road infrastructure. Vehicles that can act like a train on highways and cars on last mile, for instance. Of course, do we even need them to be self-driving to do this? Not really; basic constant-distance technology, live in the wild today, would be sufficient. Which illuminates that the blocker is political will or economics, not technology. Importantly, last-mile road capacity is very limited in cities and is the true bottleneck (there is no space to build more roads, it's not just a matter of investment), no matter who is driving or what they are driving. Hence, subways!

China is best positioned to deliver self-driving cars: social and political risks can be controlled, and technical / safety risk accepted in a calculated way (as opposed to in the court of public opinion). Additionally, they can invest in and mandate dedicated SDC infrastructure.

Still, it feels a very long way off, even for China. The driving task is much closer to AGI than people give credit. And they face the same basic physical constraints on throughput of roads that we do.

Not that I'm mad about the over hype of self-driving car programs, I think they are high positive ROI for society from a robotics research perspective alone. So pile on VCs, for the social good.

Anyways, let's build subways!

(I didn't touch on the safety aspect of SDCs but I will just leave it that they are entirely unproven on this front, all we know for sure is: 1) complex computer systems deployed in the real world have always been error prone in past – the much simpler task of automated flying still alludes us / Boeing, for example, and 2) their track record so far is not great – the fatal accident rate for SDCs is 10-100x that of humans currently, thanks to Uber)

Essays