Sunday, March 26, 2017

Uber autonomous vehicle accident and self-driving cars in general

An Uber autonomous vehicle accident on Fri Mar 24th has caused a suspension of their autonomous vehicle program for the time being.

I've always been a skeptic of the promise of autonomous vehicles. But not because of things like this - there have been relatively few autonomous vehicle accidents, all things considered.

I think that the promise of fully automated door-to-door transportation will be realized as something much more diluted. And then we'll probably collectively pretend that what we got is what we wanted all along (similar to how GPS delivered a watered-down version of the self-piloting flying car - we're not talking about the flying car part anymore).

I think we'll get some interesting and beneficial crash avoidance and safety-enhancing features as well as some new takes on cruise control coming out of the technologies involved in autonomous driving; and I think we'll get some closed-circuit autonomous vehicles that work within a defined and well-mapped/instrumented areas or road lanes in the public realm. These types of vehicles already exist in industrial settings, and existing semi-autonomous vehicles like planes work in a controlled airspace where they will only encounter other trained professionals driving ridigly-maintained vehicles.

But I also think that we'll quickly realize that automated driving only really works when you can fully rely on it to take full control, rather than requiring tentative ongoing attention from a human driver to take over at any given time. Some car companies are always working on this assumption (others are not).

One big problem with fully automated driving is that you need to be perfect and there's no room for the 80/20 type of approach that so much other automation depends on to add value, where the automation does 80% of the work and leaves 20% of the automation failures and/or work that can't be done by the automation to be highlighted and shuffled off to human workers for completion. This speaks to the vast majority of automation. It's why existing semi-automated vehicles (i.e. planes) are manned by redundant (pilot and co-pilot), highly-trained professionals.

Automated vehicles need to be consistently and overwhelmingly better than human drivers, because what you need to convince people of is that the car is a better driver than them and not just better than the average. People don't see themselves as average, and there are truly good drivers that never get themselves into "accidents" and awful drivers who leave a trail of destruction behind them.

To have any chance of success, assuming we get near to the fully-autonomous capability regardless of climate, terrain, or road condition (again, I remain a skeptic), I don't see this happening without:
  • Mandatory maintenance schedules to ensure mechanical soundness of the vehicle and operation of the autonomous equipment (again, as with existing semi-autonomous vehicles)
  • Refusal of the vehicle to operate in certain conditions (i.e. poor weather)
  • Clear rules on liability when accidents occur, not involving the human driver.
It's starting to look more and more like this autonomous vehicles need to become a fleet-based service that you use on a pay-as-you-go basis rather than something that anyone owns. Hence Uber's involvement, I suppose.

Aside, I don't know what type of successful, advanced automation people look in their day-to-day lives to as a sunny reference point when they expect autonomous vehicles to become a rapid and unqualified success. My microwave still can't cook my food to perfection. I still need to cut my own lawn. Above-grade rail is still mostly run by humans yet operates in a restrictive, controlled environment. Shouldn't those be easier nuts to crack?

And we haven't even talked about the unions yet.

Saturday, March 25, 2017

Let's just call it... 20170319

I continue to be impressed by the complexity and potential of tools like Ableton Live (and I am quite late to the game in appreciating this - having grown up with Cakewalk). I'm still quite early on the learning curve, but here's something I put together using Ableton Live and Komplete 11.