I mentioned Brad Templeton recently. I’m a regular reader and fan of his blog. He’s a smart guy who has been directly involved in autonomous car technology for a long time (advising Google, for example). He’s usually got something interesting to say on the topic. That said, he and I have a profound disagreement about the core of the technology. He believes that robotic cars must be programmed to handle situations encountered on roads as they exist today. An autonomous vehicle, he believes, needs to be able to handle itself in the exact same traffic system that humans operate in. Knowing roads and human drivers like I do, I am deeply skeptical.

In this post he emphatically asserts, "I have declared it to be the first law of robocars that you don’t change the infrastructure." In the post’s comments, responding to my skepticism, he says, "Even the most tiny of changes to infrastructure take many decades to become universal. The progress in AI is much faster. Nothing here involves strong AI." To me every one of those claims is dubious (e.g., e.g., e.g.).

To me the absurd part about this is the aversion to building infrastructure. I don’t know why the autonomous vehicle industry can not conceive of special road lanes or other accommodations to make autonomous driving much more reliable and less complex. They just can’t believe that the costs can be tolerated. Yet look at freeways. Freeways are a special purpose modal infrastructure to accommodate high velocity cars somewhat more safely. Obviously we can do this.

The example I want to focus on which proves that extensive infrastructure can be built is trains. Even more infrastructure intense are underground trains.

Rails have excellent energy saving features and trains should probably figure into a healthy balanced mixed mode transportation system. However, Brad and I both believe that many subways could possibly be replaced by subterranean roadways. He doesn’t think it’s worth digging a tunnel while I do, but if the tunnel’s dug, we agree that rail may not be optimal today. The problem with trains is that they only enjoy their efficiencies when pulling a lot of cars on a pretty static route, the longer the better. Urban subways, however, tend to prioritize flexibility and low latency.

This discussion was prompted by this article where Brad discusses some extension of a BART subway. Brad points out many important advantages to using special purpose autonomous shuttles.

  • Stations would be radically simplified for passengers. They wouldn’t even need to be underground. This would imply much cheaper stations.

  • For energy efficiency, shuttles could mostly surface instead of brake at station stops. The reverse when departing.

  • Elevators or ramps (like a parking garage) can work fine too.

  • Works fine in a dense urban area.

  • Can be integrated with other dedicated routes.

  • Flexibility of pick up and drop off points.

  • Can interface directly with other transport (trains, surface cars in parking lots, ferries, etc).

  • Powered by sliding contacts like trams or batteries as needed.

  • Vehicles involved could be cheaper than the rail steel.

I add to that these further advantages.

  • In a tunnel shuttles can do quite well with aerodynamics by drafting. Or even ventilation pressure tricks.

  • Different capacities are possible to fine tune logistics and rider preferences.

  • Better flexible routing would reduce stopping/starting energy.

  • Less stopping can improve passenger travel times.

  • Coordinated pullouts allow bidirectional travel on one bore.

  • The "road" minimally would only need to be two strips.

  • Intersections and changing route is relatively simple.

  • Unlike trains they can employ underground roundabouts (e.g. Vallavik Tunnel).

  • Tunnel fires are always bad news, but non-preferential bi-directional drive could help orderly evacuation.

  • Tunnel fires can be fought with unmanned fire trucks.

  • Tunnel fires won’t be gasoline fires.

  • Better emergency braking. More responsive than trains to people or objects in the wrong place.

  • New infrastructure could be explored without disturbing existing systems.

  • Doesn’t block main conduit during loading. This could be nice for freight or handicapped access.

  • Can run a noisy high throughput transportation line through a "quiet neighborhood".

  • Efficiency improves when load is more consistent, not mostly empty or mostly full as with current mass transit.

  • If the energy savings of rail is still compelling, the shuttles could be designed to roll on rails also and they could drive themselves from the pick up to the tracks, engage the train wheels, and zoom away. They could even hook on to each other as an ad hoc multi-car train.

  • It might be a good precedent to send our high energy transportation underground where it doesn’t kill people willy nilly.

  • It would help to smoothly transition to the autonomous vehicle future that seems inevitable at some point.

That last point is more important than it might at first seem. Brad advocates rubbing the machine learning lamp until the AI genie is smart enough to deal with, say, me on my bicycle. Although that’s unlikely to ever happen comprehensively (as the recent fatality shows), there is a more poignant aspect to that vision. Once autonomous vehicles are successful, facilitated by superhuman AI (endorsed by Brad and the entire industry) or dedicated infrastructure (me), it is clear that superhuman AI definitely will not be necessary. This is because the only serious problems that remain for autonomous vehicles are the human drivers they are trying to eradicate. I say if the AI helps, great, but rather than wait another half-century to find out that human level intelligence isn’t just around the corner, let’s do the obvious stuff, isolated from human randomness, and start eliminating driving in a way we know we can! Obviously people are willing to build things like subways. Let’s make that investment more effective with the smartest technology possible.

Here is an inspirational rough sketch I made of what this idea could sort of look like (if the world was designed by someone who needed Blender practice and reduced to 32 colors for your downloading convenience).

Xed’s Concept Sketch

This concept provides an illustrative look at the advantages of autonomous vehicles without worrying about cost, integration, safety, consumer acceptance of sharing, loading areas, parking, trolley problems, zoning, noise and disturbance of 24/7 operation, fare payment systems, liability, "taking away jobs", etc. — the subway already had to deal with all that! It’s an ideal case to objectively compare one big complicated expensive transportation system to another. I believe the case for autonomous shuttles over rail is compelling but what is truly compelling is at least opening our minds to the possibility.

UPDATE 2018-03-27

I just noticed this project from The Boring Company. Seems like a good idea.

UPDATE 2020-08-19

Brad just wrote another excellent article about this topic. He’s obviously been thinking about this and has added some more compelling features to this concept.