Welcome to Afternoon Latte! This is a series I am trying out. It is intended to be a little something interesting and not to terribly deep to get you through the end of the day and into the evening! We'll see if I can keep finding light and frothy things to talk about but for now, take a sip, it is just the right temperature!
Okay, time for a little evolutionary fun. This link will take you to the BoxCar2D site.
At this site you can watch as cars run through an evolutionary algorithm. It starts with a randomly generated stet of triangles and wheels and test them for “fitness” by running them over a variable two dimensional track the further that car gets the more evolutionary fit it is considered.
After 20 cars the two fittest are chosen to be the “parents” for the next generation of cars. This should mean that the cars improve in fitness over time. However they don’t just conserve the traits of the previous generation, you can control the mutation rate raising or lowering it.
This illustrates very well that mutation is not always good or bad. Many of the new mutant cars basically fall on their heads and die without getting anywhere on the fitness test. Others seem to have all the traits needed but as they hit the various challenges of the track then can be stopped by not having enough traction or by flipping over on their heads and dying.
The inventor of this little time waster also set a boundary condition so that in any given generation a car can only go so far. It is an average of the total amount of fitness of the previous generation, that is what the number in the parentheses is.
You also have the ability to control the maximum number of wheels and how frequently that maximum is reached.
All in all it is great way to show how evolution happens, and what happens when there are more or less mutations in a single generation of a population. Over time the cars do get better but it is an up and down process instead of a single rising line that so many people think of evolution as. If you let it run for a while you will even see the “regression to norm” where extreme progress is very rarely kept, but rather after a spike the averages will tend to return to the same range. There are also plenty of “unfit” mutations just like in real life.
Go there and give it at try. It is endlessly diverting and you will find that some of the cars you think are going to be winners turn out to be less fit than the oddballs. Be sure to let it run for a while, and do play with the controls for mutation rate!
The floor (or in this case the road) is yours!