How Do You Solve a Problem Like 100k Uncoordinated Driverless Cars? (opens in new tab)

> They don’t drive like humans, they don’t negotiate intersections like humans, and they’re going to interact with human drivers and pedestrians in unexpected ways.

Well, if you can't make a car that behaves like a good human driver, then maybe you shouldn't be making driverless cars? Otherwise, instead of reducing accidents, you might increase them (at least initially, before humans adapt). Self driving cars that can't interoperate with humans using existing infrastructure are basically useless - infrastructure lifetimes are orders of magnitude longer than technology life cycles. Re-planning cities for driverless cars now will make you regret it when the technology matures.

Also, while centralized planning might fly in current day '1984' London, I sure hope this fails at least in EU, where they claim to value privacy. Nothing about self driving cars requires either centralized reporting or even V2V - if you can detect a pedestrian behind a tree, you sure as hell can measure the speed of the car ahead.

Kinda feels like an advertisement for the operating system mentioned multiple times in the post. But, how do you solve the problem of many uncoordinated driverless cars?

You don't. What you do is avoid the problem by defining protocols and regulations. Then by updating the infrastructure to allow the vehicles to operate in sync but also out of the way. Self driving cars, at least the first generation or two, will require their own environment. Due to the multitude of variables to handle. Insurance, public awareness and education, safety, etc.

In the same way that dumb old cars required (and still do!) new regulations, protocols, and infrastructure amongst other things.

In terms of technology, how do you coordinate all of this cars? Well, that's one of the things I'm trying to find out. There are many, many papers out there that have been attacking the problem for decades and the small subset that I have managed to read agree on one thing: The infrastructure needs to be updated for self driving cars to operate. Not only roads, but pretty much all city planning aspects. Now, you think San Francisco is making a fuss about not building new housing. Imagine how it will respond to redoing everything. Now imagine that at a state level.

It's not a problem solved purely by technology, but through consistent trial and error of multiple moving parts.

So what I envision is a many layered network of signals between cars. You have a pretty wide diversity in the importance and latency requirements of various communications between the vehicles regardless of the medium used.

Something like "Stopping" or "Everyone Stop, accident" need to hit the wire as quickly as possible. Maybe order the message to the Emergency Stop hits the wire first and any further message information after. That way the message can bypass most of the communications stack in the receiving cars.

Less important information like "Deer on shoulder of road" can use much slower and more latent communications.

Distance is also going to play a part in how the cars communicate. If the car can see tons of low latency high priority messages then it will be more problematic to develop. If the low latency physical layer can only see maybe 100 ft around its self because it is low strength radio or IR based, then all the better. Loud wide area communications can go through a far more latent filtering system.

All of this also doesn't take into account that these systems should degrade gracefully. If the low latency portion fails, the high latency system should still see something and react.

If the cars know how drive themselves you don't have to coordinate anything. There are things we can do to optimize traffic and whatnot by allowing these autonomous vehicles to communicate, but if a car can handle driving around distracted humans, it can handle driving around other cars.

Given how car media consoles are so different and poorly coded, this is a worrying point! Would be good to make a protocol for cars to communicate intentions to each other. Xbee mesh networking maybe!

There perhaps should be a highly regulated frequency that is used for "Emergency - stopping dead now". With some localized physical coordinates (GPS possibly being the wrong granularity) so all within a certain radius know to immediately begin slowing, the closer the more radically.

Blockchains allow transfer of value even between parties without contracts or deals or even trust, by exchange of a rare token. For eg Cars from different companies may negotiate among themselves using bitcoin, this was never possible earlier because for cars to exchange tokens theyd have to be under some central authority or some contract. Bitcoin provides the ability for signing contracts on the fly.. Another eg would be a car stuck somewhere opens a rescue request on a website. An automated crane bids and goes and rescues the car and goes away taking the bitcoins. For this to exist today some deal is needed beforehand.