Someone mentioned another evolution program earlier and it rings a bell. It's one where you load preset unicellular organisms with genetic code determining their behaviour, into a virtual environment where they compete for survival with other loaded organisms. They mutate over time, with stronger mutations being more successful etc. It was called genebots or cybots or something along those lines.

yeah that video shows something like this. It makes much more sense.

Yeah, I gave up on it to. The problem as I've identified previously, is that the effective mutation rate is much higher than whatever you select on the slider, so there's very little fine-detail control. Perhaps the bigger problem is that the degree of the mutations possible means most of the time any particular mutation will be detrimental, and this coupled with the speed to complete the course (seriously needs a 4x or 8x speed mode) means it takes many many generations to find good solutions. Lastly, the randomly generated terrain means some courses are going to be very very difficult to solve, or generated in such a way that while a car can easily get up to score 180, getting up to 190 is almost impossible because the type of design required for that wouldn't have been able to get up to the 180 mark in the first place. If it were possible to design your own track, or at the very least specify a seed for it, it'd be a lot better.

New version: http://www.boxcar2d.com/

This time with eugenics!

I found the new version produced a lot more failures than the old one, mainly due to the wheel mutation. However, my observation clashes with my results, which show a HUGE exponential growth in the cars' ability. So I'm a little stumped.

An improvement, and he describe the algorithms used for generating the cars. I was pretty close with my theories, although there are only 22 variables not 36 in version 1, but I was right about how the mutation rate is applied (stupidly).

In this one, though, you can have up to 8 wheels, and the wheels can also been much larger and apparently can vary their speed as well. So the best designs will probably end up being cars that have ginormous wheels where the bodies are fully immersed in the wheels themselves, so that it can't hit terrain and stop it.

In generation 6, I've had two cars get to 284, which is much farther than my previous best (212). The cars with heaps of wheels just kill the slope at 200-210, and the jagged W shapes don't pose much threat either. I've found that with cars with multiple wheels they tend to do a lot of flips in the air while they're driving along, whereas previously a flip would generally halt the car right there, these ones can keep going (sometimes faster after the flip).

Post has been edited 1 time(s), last time on Jan 28 2011, 10:28 pm by Lanthanide.

Wow, that just got really awesome. Upping the number of wheels helps a lot - it's what I thought was limiting the ability of cars to handle rough terrain last time.

At the moment I'm thinking the best design would be one with two large central driving wheels, and two smaller guide wheels raised away from the road surface at either end, to prevent flips, provide shock absorbtion and a bit of extra power to get up hills.

To bad you can't design a starter car and allow mutations off that design.

This one eventually got to 421.5. By far the biggest difference seems to be the much increased torque, and increased wheel size. These wheels spin really fast, so a lot of the track up to 300 or so it's flying through the air as it goes up hills, and tends to glide over the top of the W and M jaggy bits, both from it's built up speed and increased traction when the wheels just touch the top of them, and of course the wheel size stops it dropping too far and getting stuck.

Attachments:

I hereby call out the unicycle challenge. Who can beat my breed?


As a side note, the part where the red line fluctuates a lot I set some mutation. Left it running for too long cause I went away and it really degraded my performance for many generations to come (see straight red part) when I finally added a bit of mutation I got a huge jump I didn't even thought were possible for unicycles. Quite entertaining I have to admit.

But wouldn't that be...
intelligent design?

I made use of the up/down buttons to get me a consistent 421.5 population, as pictured. I set mutation to 1%, and within 5 generations all of the 421's had gone and the best score I was getting was around 220 or so, with lots worse than that.

I think my winning design came from 4 basic factors: huge wheels, very high torque on all of the wheels, low slung body that was immersed in the wheels (never flipped) and wheels that were fairly wide apart so it could easily glide over jagged bits - they only look close due to the size.

In the screenshot above, the red line has actually gone up off the top of the app, and once the whole population started hitting 421.5, so did the black line.

Yeah similar for my long time - multi wheel approach. I got just over 500 max score but then there's a small, but extremely steep pit (~200% pitch) situated on an already steep slope. The car is pretty slow on that slope and then just barely manages to get the front wheel over the pit making it even slower, and then the back wheels just fall into the pit and the car is stuck because it's not strong enough to climb the steep pit. I don't know if that's intentional or not but I think engine power should not be an issue in this simulation.

Remember when this hill would've killed the cars? Now the cars kill the hill

Easy... But nobody tried unicycles? They work surprisingly well and it's funny to watch what strategies they use go move on.

They got an editor now. And breakable cars.

I'm not so sure if breaking cars are a wise decision.... However it is lulzy to see the speedy cars constantly wrecking themselves.

Sometimes a car can crash from a nasty bump or something, break part of itself off, then resume riding even better than before. It potentially lends a new layer to the strategy, if you think about it.