The Autonomous Tricycle

Earlier this year I described how autonomous vehicles could be usefully deployed today with no further improvements in technology. And in another article I proposed a completely different way that autonomous vehicles could be usefully deployed with today’s technology. Not only could they be deployed with today’s technology, they could be deployed with 2006 technology. Not only could they be deployed, they should be deployed.

To press the point, I have yet another idea for how autonomous vehicles can and should be deployed today using technology that was available in 2006. I can summarize the strategy in 2 words: minimize energy. Let’s look at what that means and why it might be a superior solution to the ones typically being considered.

When I’m riding my bicycle along the beach boardwalk and mixed use recreational paths near where I live I am extremely aware of my responsibility and liability. If a drunk person jumps out and collides with me knocking me off my bike, it’s easy to know whose fault it is. It is my fault. It is always the cyclist’s fault when interacting with pedestrians. (Just as it is always the cars' fault when interacting with cyclists and pedestrians - but that’s another topic to be argued at another time.)

Here’s a frightening situation I was in over 25 years ago that I still vividly remember now. I was cycling next to a line of cars mostly stuck in dense traffic. Just as they ordinarily pass me in their perceived space, I was now passing them in mine (legal for motorcycles to do in California). Suddenly the car I was next to and overtaking slammed on its brakes. As I rolled in front of it I realized what was going on. There was an adult crouched down on the far side of the road encouraging a small child on the near side to run across. Although I have good visibility, I didn’t see the adult because they weren’t standing and the car I was passing was in the way. I didn’t see the child because he was in between two parked cars and short. The adult was telling the child to go because they couldn’t see me (and they were being stupid). Well, that kid can thank the fact that on that day I had superhuman powers of bike control. As he sprinted right toward my back wheel from point blank range, my attention focused only on the child and keeping my bike from making contact. I don’t know how I did it, but I unloaded an extra few kilowatts and dropped the back wheel fast enough that he just missed me. I even somehow managed to not wreck myself.

Here’s what’s important about that story now—in 250,000 km of cycling I’ve almost been slain dozens of times but that is the only serious close call I’ve ever had where I was culpable. I could have killed that child but I think it’s more likely that I would have given him a nasty mauling. I’m not even going to say that I’ve always not been stupid when cycling around pedestrians. The important thing to think about here is that no matter whose fault it is, the most severe collision I can think of between a cyclist and a pedestrian is much pleasanter than the most mild collision I can think of between a car (at speed) and a human.

What does this have to do with autonomous vehicles? Currently the leaders in the AV field are selling Teslas, Mercedes, Cadillacs, etc., huge massive cars with enormous power. What if we reconsidered the list of problems involved in designing an autonomous vehicle by limiting the vehicle to use the same order of magnitude of energy as I do on a bicycle. (For reference my 1 hour PR is about the range and speed of recent electric bicycles, around 40km.)

With that inverted perspective suddenly the situation is radically different. Sure you could hit some person or dog or other unexpected soft thing, but the severity would be so much less that it might not need to predominate all other considerations. With limited energy you won’t be going 80mph. You won’t be towing a boat. But I can imagine a useful very small light vehicle that can travel at bike speeds. Autonomously.

I’ve been thinking lately about building my own autonomous vehicle. Rather than dismantle and modify a crushingly expensive real car only to find that it is an intractable menace to humans, I’ve been thinking about a tricycle. It could be a dorky looking thing someone’s grandmother might ride or something fancier. But the idea would be to combine it with an electric drill (essentially) and a laptop. Ultimately the laptop could do double duty autonomously driving the rig while entertaining you.

Where could this be used? Why would this make sense? Well, it’s kind of a tricky gray area right now. Although I rode one in the Swiss Alps in 2001, electric bicycles are just recently a major thing. I have mixed feelings about this. Another word for a motorized two wheeled vehicle is "motorcycle". But that’s exactly what I’m arguing here - it’s good to explore different regimes of power and weight physics than we’re used to based on old technology constraints.

If you believe in electric bicycles as bicycles, then the nation of Holland could be transformed today into an autonomous vehicle utopia based on the fact that they have already properly insulated humans from their dangerous car infrastructure.

The concept is this. You should be able to climb on your autonomous tricycle and although cool people will be laughing at you the whole way, smart people will notice that you’re getting work done on your commute. Maybe your commute would take longer, but I’ve always argued that focusing on an obnoxious task for 10 minutes is much more wasteful of one’s precious time than doing whatever you want in a confined space for 20 minutes.

Folks, let me tell you something as someone whose life depends on being an exceptionally keen observer of this fact: the killer app today is being able to text while driving. It is not just figurative; people will literally die to be able to do this.

Although you probably were thinking autonomous vehicles were going to look like this…

…I’m suggesting keeping an open mind to having them look like this.

One of the added bonuses of a low-lethality approach is that it bypasses regulatory challenges. It is approachable by real entrepreneurs, not just lucky rich guys. It can be augmented with solar power and (gasp) human power. It’s a massive improvement on the traditional approach in terms of energy usage and pollution. It optimizes for the absurd problem of most solitary commuters driving a wasted capacity for 5. It is easy to correct if it gets into difficult situations (ahem). It becomes a better idea the cheaper batteries, computers, and sensing hardware become. Building custom infrastructure for such low energy vehicles is orders of magnitude cheaper and easier. Although this would work magnificently in Holland, it actually has the greatest potential at the opposite end of the infrastructure spectrum in the third world. The absurdly rich have had autonomous vehicles for millennia; a low barrier to entry opens the possibility to poor people, motivated people.

The goal is not to develop a car like KITT or Optimus Prime. We do not need cars to be so human that we can be friends with them. I sometimes feel like that’s what AI researchers are going to be stuck working on forever. What are the real goals?

What is not on that list is "look like a normal car". Instead of taking the physics profile of a 2000kg car and wondering how we can make it drive like Roy Batty, why not start with the physics profile of a bicycle and ask what is required to achieve the goals of autonomous driving?

UPDATE 2019-07-31

Here’s a weak article about a vague Chinese attempt at an autonomous "bicycle" (i.e. a motorcycle). How can this be news? Has everyone forgotten Anthony Levandowski and Ghostrider?