Comment I was a little harsh (Score 1) 824
on the coders. I apologize. I've just had serious issues with traffic control in the past, and have encountered significant resistance. However, this does not come from the coders themselves. I do not apologize to the unimaginative buffoons though. If you've ever dealt with the people in charge of state offices, then you probably know what I'm talking about.
I understand the appeal of classical controls. I mean, everything is exact, and there is no uncertainty. You have all of your models and you can design a controller that has a rise time of x, settling time of y, overshoot of z, etc. It's very precise, and appeals to the extremely logical, mathematical nature of most engineers.
However, how accurate is that model? The model is linear, but is it a linear system? No, you have to linearize it. That linearization is only good over a certain range, and after that it fails. A linear model says I can pump 100 A through a 1 ohm resistor, but I know better.
There's also the issue of noise, but the main problem with traffic control is the model. People are highly nonlinear, and their behavior is not easily modeled. Deriving a transfer function for the driving habits of city full of people is an impossible task. The best you can do is come up with probabilities and go from there. Now, I don't know about you, but I find the math involved ugly and cumbersome.
So why not derive a set of fuzzy rules? A simple case is classifying traffic as light, medium, or heavy, which would correspond to short, medium, and long green phases. The math of membership functions is extremely straightforward. Of course, this is an extremely simple example, but intersection control is still fairly intuitive. Anyone sitting at a red light at 3 am can see that nobody's coming, so there's no reason they shouldn't have a green light. With fuzzy logic, you use the same kind of reasoning, and come up with a much more efficient traffic controller. There are some instances where an imprecise approach can lead to much better result, and this is one of them.
I understand the appeal of classical controls. I mean, everything is exact, and there is no uncertainty. You have all of your models and you can design a controller that has a rise time of x, settling time of y, overshoot of z, etc. It's very precise, and appeals to the extremely logical, mathematical nature of most engineers.
However, how accurate is that model? The model is linear, but is it a linear system? No, you have to linearize it. That linearization is only good over a certain range, and after that it fails. A linear model says I can pump 100 A through a 1 ohm resistor, but I know better.
There's also the issue of noise, but the main problem with traffic control is the model. People are highly nonlinear, and their behavior is not easily modeled. Deriving a transfer function for the driving habits of city full of people is an impossible task. The best you can do is come up with probabilities and go from there. Now, I don't know about you, but I find the math involved ugly and cumbersome.
So why not derive a set of fuzzy rules? A simple case is classifying traffic as light, medium, or heavy, which would correspond to short, medium, and long green phases. The math of membership functions is extremely straightforward. Of course, this is an extremely simple example, but intersection control is still fairly intuitive. Anyone sitting at a red light at 3 am can see that nobody's coming, so there's no reason they shouldn't have a green light. With fuzzy logic, you use the same kind of reasoning, and come up with a much more efficient traffic controller. There are some instances where an imprecise approach can lead to much better result, and this is one of them.
