Building Autonomous Road At Home
Autonomous Vehicles (AV) have been discussed widely, so much so that I wouldn't consider myself even to be a beginner. The subject has many technological aspects (to say the least), many of which I have no clue about. There is one thing I can say, however - I think that the mainstream discipline of how to control individual cars is, with all due respect, wrong.
Let me first make it clear that when I say 'how to control', I mean who should make decisions about this. AV manufacturers are following the mainstream discipline of making this decision within the car itself. This means that you need to invest a great deal of resources into placing an extensive number of top-notch sensors inside a car in order for it to collect enough information so that it can make a good guess at what to do next.
For my part, I believe that we could achieve a safer, more cost-effective and easier solution if the decision was not made for a single car - but for all of them at once. And this is (largely) why:
Aspects of consideration
Object localization is at the heart of the controlling decision mechanism. Although self-localization is a tough problem (with centimeters precision) - one could argue that by using (actively helping) infrastructure, this could be much easier, more reliable, and cost-effective.
Coordination is another aspect of the controlling entity. True, you can coordinate a number of cars by grouping them into a fleet, which has numerous researched solutions. However, this solution is best suited for routine driving or highways. During times of need, vehicles might be unable to coordinate an optimal course of action under specific circumstances. By placing the decision entity outside of the vehicles and centralizing it, it may be easier to recognize hazardous situations and act sooner.
This means that rather than passing warnings about trouble down the road and then making similar decisions in all the vehicles in the proximity, as each car needs to respond separately to the warning, the warning is intercepted once, the decision is made once, and the control instructions are broadcast simultaneously.
Yet what about computation? Somewhere, something has to compute the future actions of all these vehicles. As it happens, we have just as much computing power as we need. Having centralized decision-making does not imply centralizing computing. Why not use the computing power of the cars themselves for this task? Decisions are centralized - computations are not. Just to clarify - cars aren't making decisions for themselves - they're collaborating in computing the centralized decision.
Essentially, the suggested system consists of a giant robot with countless branches, each of them participating in the robot's computing and communication.
What do I want to do?
I'd like to build this robot! Well, at least, a pico-tiny version of it with RC cars. This is not (yet) a money-where-your-mouth-is thing - It's more of a challenge with a few very interesting checkpoints that I'd like to pass.
As of now, this is how I view the project’s scheme:
- The car should be controlled by a computer - MCU or a PC, it doesn't matter for now. For this, I decided to use technic Lego - a long lasting childhood hobby of mine. I'm talking about the Ford-150 Raptor set, of course.
- To control an RC car (or something similar), one needs to know where it is. So I need to come up with a scheme to localize the car.
- Once we have our building blocks, we can now develop the mini robot's software - to control it!
Updated Content Table:
