A couple of years ago I was visiting my dad and I was introduced to this toaster.


This is a fancy and somewhat expensive toaster. Just look at that fancy digital display! I don’t know about you but for me and many others toast is a breakfast thing. Sometimes I get up and make toast while other people are still sleeping. Imagine my surprise when this toaster announced the readiness of my finished toast with a shriek like a chimpanzee being disolved in acid. It scared the hell out of me! I wondered, what moron designed a breakfast food preparation appliance to literally sound like an imminent train wreck?

This would not do. Annoyingly the toaster was assembled with special security screws. That just made me even more committed. First I had to make a tool to disassemble the thing.


Having accomplished that I was able to open it up.


And here’s the obnoxious source of all the fuss.


This spec sheet shows a very similar 1205 buzzer producing 85dBA. This means that it is similar to a dump truck driving by. I was able to disable the buzzer, reassemble, and finally make toast in blessed silence.

That was a toaster problem.

It was not the toaster problem.

The toaster problem is a shorthand phrase I use when discussing autonomous vehicle technology or any futuristic technology that is heavily dependent on artificial intelligence that hasn’t quite been invented or perfected. The toaster problem is that toasters can not reliably toast a piece of bread. All toasters I know about may be able to be set up to make one piece of toast satisfactorily, but if you scale that to 9 pieces of toast, maybe on a cold morning or a hot day, well, if you want your toast perfect, you, a human, will have to keep an eye on it. No big deal usually, but the important point is this: if we can’t have autonomous toast toasting machines, how the hell are we going to have automatic machines that can perform double lane highway merges?

I focus on toasters because the technology is so banal and stupid. Not only that, but I personally can envision a solution. I believe that I could set up a camera to watch the progress of my bread toasting and with enough iterations, I could train a machine learning algorithm to produce a "toastedness" score which would actually be related to the toast you were wanting regardless of how hot the coils were at the start of the process. I’m a wee bit surprised that KitchenAid doesn’t have such a thing for $1000.

But why is this a useful phrase? It’s important because there are tons of things that are unsolved toaster problems. For example, trains have drivers. I happen to know that some trains do not and this industry is working hard on their toaster problems, but still — if 2d cars are to drive themselves, shouldn’t we be seeing 1d trains do it first?

Another example being worked on are floor cleaning machines that can operate with or without a driver. This is actually the first (2d) autonomous vehicle I ever got to ride.


A couple weeks ago I spent twenty minutes being actively driven around O’Hare airport on the ground in a fancy fleet-managed airplane. I would guess that taxiing in such a controlled environment is a toaster problem even if full gate to gate autonomous flying turns out more complex. Certainly ground support vehicles in tightly controlled airports could be easier than general passenger car transportation. Oxbotica is working on that toaster problem now.

Sometimes it can be tricky to place a technology concept on the toaster spectrum. Is an automated battleship easier than a unoccupied robocar that safely doesn’t kill cyclists? That probably depends on many complicated factors but that idea is currently more than science fiction. How about a car that can race off-road through the Mojave desert? It’s not obvious.

Certainly partial technological progress on the way to fully autonomous passenger cars implies many useful intermediate technologies. Is a car that can park itself essential before a car that can do everything that I can do as a driver? I’d say so. All the modern ADAS features are toaster problems solved.

I think one of the biggest mistakes made by a lot of autonomous car teams is to neglect the toaster problems. For example, Kitty Hawk is aiming for flying autonomous cars. I suppose if you’re ok eating burned toast, you might as well dream big.

UPDATE 2018-07-04

Car alarms. That’s way way lower than toast. And, a very similar stupid problem, the obnoxious klaxons of garbage trucks and backhoes in reverse. There will be no meaningful autonomous cars while these things exist.

UPDATE 2018-07-12

The Economist covers some projects that may literally be working on the toaster problem as part of a grander robotic chef.

Also there can be toaster problems in other areas too. For example, the fact that I can’t live under the surface of the ocean or in central Greenland is a toaster problem for people wanting to colonize Mars.

UPDATE 2020-01-016

Did Chris Atkeson of Carnegie Mellon University’s Robotics Institute read this? Check out the end of this video at about 19m22s where he points out how overly comprehensive Knight Rider’s KITT was. He says, "I think it’s much more realistic that we’re going to have robots that are pretty good at a limited number of things, and gradually get better and better. Rather than we go instantaneously from toaster-level intelligence to it’s as good as human." Indeed.