…and now it's sitting in another tab waiting to reveal to me its secrets.
I'm not even stoned.
Best goddamn screensaver-that-isn't-a-screensaver since ALife back in Win95 days.
This drunk guy made it way further than he should have - about 100meters. I am finding myself naming them and imagining personalities. "Oh BIG PAPIIII! Make it over that hill. You can do it!"
I think after every race the winning design is the basis for a spawn of new mutations. If the floor is fixed, you'll get closer and closer to getting the best design for that floor. This is so cool. Also yea, it's 4:46, do I really need to be awake any longer? Now I have to. edit: I didn't realize you could scroll down the page and read about the controls and genome: - Shape: (8 genes, 1 per vertex) - Wheel size: (2 genes, 1 per wheel) - Wheel position: (2 genes, 1 per wheel) - Wheel density: (2 genes, 1 per wheel) darker wheels mean denser wheels Wheel density! That's wild.I regret clicking this link as I intended to head to bed.
The genome consists of:
I get that, but does the winning car get copied exactly, or does it also mutate? Do the cars that don't make it far get completely replaced? Also, Battleship was a stupid premise for a movie, but if that got made, do you think that adding a Tron -like element to this would increase its chances of becoming a movie?
What Tron-like element are you referring to? Because it's already a race of sorts. I don't know the exact formula, I'm about to leave browser open for an entire day and come back to report.
Reminds me of this old post by mike titled Evolving a Wind Turbine. Cool stuff.
I love watching these. It really brings ideas about evolution to light, especially that the design that takes over is not necessarily the optimal design, but it was "locally optimal" in that it is better than all the designs that are small variations of it. It's interesting to let the designs evolve for 10 or 15 generations then crank up the mutation percentage to 50% or so for a couple of generations, then tune it back down to 5%. This shakes up the environment and can possibly stimulate better designs that never got a chance to evolve. It shows how important major environmental shifts are for the long-term health of a species (maybe!) When I used to teach university courses, my students would sit in groups, get comfortable, and then half-way though the term I'd use a simple algorithm to switch people around to new groups with no common factors. There would lots of complaining but people would settle in. After a few weeks, I'd let them sit wherever so they could reunite with friends if they like. I thought of it as creating a stable environment, letting it grow, then dropping an asteroid on the planet and letting it restabilize -- a mini "ideal evolution environment" if you will. It's cool to see vestiges of structures that were useful at one point and are no longer useful. I don't really see them in this simulation but there's probably examples to be found.
Red Queens (Funny -- google's 2nd item was this: http://hubski.com/pub?id=22374 )
Sweet! Do I dare keep this on all day at work... I think so. I'll report back when I've created a jet that rockets all the other cars.
Lol... and of course I completely forgot I left this running. But here are the results at a 10% mutation and 9 hours to work at it: The designs are definitely more interesting than when I left this morning. The sad part though is that a lot of the cars being spawned now still can't climb the same steep slope in the middle of the race. I can see from the records though that a number of cars had gotten past that slope while I was gone. Genetic algorithms are sweet and if the problem set is well formed can be very effective. This is a really interesting application. Thanks for the share kleinbl00
Just for completeness here is my winner. I figured out how to replay your best car. Unique properties that I noticed about this car. It has big wheels able to get it over the steepest hills. It's very difficult for it to get stuck when it rolls over. In fact it cleared many of the hills by repeatedly going at it, flipping over, regaining speed, and ultimately climbing it.
I left it running over night, which got me up to about 750 generations. I watched for a few generations then went to eat breakfast. When I came back, it appears generation 822 "won" the game, so I watched the replay. At about 220 meters, the floor generation stopped but the car kept going, falling rather. At a height of -2920 meters it stopped altogether. I've watched about 250 generations, and it seems that luck has as much as anything to do with the performance of any given car. Sometimes a car will make it farther than it's elite clone will the very next generation. In fact, clones seem to get stuck rather often, allowing possibly inferior cars to finish well beyond them. When this happens, whatever "good genes" that clone had are lost, and it often takes many generations for a car to again take the top spot.
These look very similar to what all of my cars looked like at ~750 generations. Throughout the first 150-200 generations, the best cars tended to have a large front wheel and a small back wheel but this must change somewhere in the 500 generations in between.
It was quickly fairly obvious that the ideal shape for one of these things is where the shape doesn't protrude past the wheels. After about 40-something generations and some fiddling with the settings, one finally popped up. A version of this one is still top dog at gen112. I'm also noticing a huge improvement in car performance when lowering the mutation size in particular.