There has been a lot written recently about (semi) Autonomous Vehicles (SAVs) aka Self driving cars. Especially yesterday following the Tesla announcement, see also this NPR report.

Self driving cars have long been held out as a solution. They arrive when you need them, they take you to your destination at regulated speeds, they can adjust to traffic congestion, and collisions, road work etc. Once you’ve arrived, they disappear not requiring a parking space either in a building, or on the street. Nirvana.

Proposing self driving cars as a solution to traffic congestion, where the congestion is caused by constraints is simply a nonsense. They are a first world solution, to a first world problem. “Why can’t I text/read/sleep while I’m stuck in traffic?”

There has long been discussion among the urban transportation advocates, while they promise to reduce accidents and better manage traffic density and improve capacity through stable and reduced “headway” (the distance between cars). Anyone who proposes they can solve traffic congestion without understanding the constraints and capacity issues, just doesn’t understand the challenge. This Forbes article screams “Self-Driving Cars Would Slash Traffic, End Street Parking” yet aside from the obvious nod to headway and predictability doesn’t address the problem. This Qoura discussion covers many of the points

Here is a response I wrote yesterday on the issue to the #ATXRail mailing list. It refers to an earlier post that discussed in some technical detail the problems of scheduling and capacity.

Again, I find myself being the naysayer, Roger Cauvin made some extremely salient and possibly too technical points about any transit system design, but especially as it relates to individual transit options such as autonomous cars. The problem that most simply don’t understand is the scheduling and availability of these cars at scale.

First, whats scale? How many would be needed to make a real difference? Scale certainly isn’t a hundred, it’s not a thousand, it’s probably 10,000 maybe more. Second, lets assume these are free movement vehicles, they don’t run on track or are constrained by guide rails. Third, lets assume that they are capable of transporting 2-3 max. but the assumed capacity is one passenger. Fourth, lets also assume these are not individually owned vehicles, or at least if they are, they are available ala Uber/Lyft for other people to book. Fifth, lets assume they are electric and capable of driving themselves to charging stations where they either connect or are connected to a recharge point.

Now we have the basics of your capacity based system and you can start working on its constraints. What you don’t have is any real clue about the usage patterns, how they’ll be used, who by, for what, at what time and so on. Until you can produce even a first pass for this you cannot make any assumption that autonomous vehicles can solve anything except perhaps more consistent traffic flow, and improved emissions. Anyone who claims autonomous cars can solve anything without this data is simply blowing smoke and must be pushed back on.

They are great for private vehicle journeys. I have a Mercedes that will already follow a GPS journey, has variable cruise control that will slow down and speed up the car based on the MAX speed of the vehicle a set distance in front and will bring the car to a stop based on either breaking vehicles or an obstruction in front; it pretty much does everything except steer, but it vibrates the steering column to let you know when to turn and when you’ve are moving out of a lane etc. This type of vehicle which we’ll see for all new vehicles in the next 5-7 years, if not before, will much better regulate driving.

They’ll minimize the concertina effect of drivers speeding up, jumping lanes and breaking too hard which will make congested roads flow much more smoothly. In practice they can also stop people jumping traffic lights, travelling faster than the legal speed limit and other traffic law infringements.

This begs question will drivers allow this to happen, and can insurance companies be convinced to drive the adoption of it?

So assuming all that gets done, autonomous vehicles will become totally acceptable and usable. And then we’ll have the scheduling issues that Roger alluded to. They are real.

Having self driving parks valet park themselves is cool, the question is where? At scale, say 5,000, you need at least say 2,000 spaces, and over night, probably 4,000. Lets those spaces also need to be charging points. Someone has to invest in building and powering those points. The “free” market will(really?). Then all you have to go is schedule cars between where they are, the users, the users destinations, and the parking spaces. When the primary destination is still downtown Austin, you have some massive constraints, not magical relief.

Although I don’t work in that area now, I work on computer systems scheduling for 15-years, specializing on at-scale systems. These were airline, banking and transaction systems. That had very similar constraints in so much as they all used run one one or two mainframe systems. I was the lead architect for a system we spent more than 2-years modelling an internet banking system that eventually successfully supported 900,000 concurrent users on two servers. I helped fixed the design and scheduling for a single system that supported 23,000 concurrent users, a record at the time. Before you all say but yeah cloud computing has changed all that, it really hasn’t.

The design and use pattern considerations Roger discussed are key. If you had 100 destinations you’d have removed some of the major bottlenecks, but we don’t. We have the Central Business District. It’s a major constraint and getting people in cars in and out of it no matter how automated the cars are will have the same fundamental problems and constraints. Anyone who says otherwise really doesn’t understand the problem.

It’s true, as automated cars become standard features of our cities, it will be easier and easier to write the scheduling rules to make them work within the unique constraints each city has. Austin will have more than many cities constraints because of the sprawl, because of the access road problems, and especially if we have not viable alternatives.

On the “self driving” Mercedes observation, I had never thought of trying this.

Update: 1:23 added in Qoura link.