Evolving 2D Cars

[Recusant]

I suppose I could stretch a point and post this in the "Science" section, but it seems more just entertainment to me.  A brief description: "Watch genetically-evolving cars race in 2D" | BoingBoing

. . . And a direct link to the page: Genetic Cars 2

[xSilverPhinx]

 

I suppose I can't expect to get a Lamborghini-like vehicle in this simulation.

My evolving car creatures are stuck.    For many generations they haven't been able to get past some sloping terrain.

[Recusant]

It's possible to vary the mutation rate as well as some of the other parameters. Using the default settings, I let the program run until one little monster evolved that was able to complete the course. It took only 50 generations, but your mileage may vary.

[xSilverPhinx]

I let mine run using the default settings for 63 generations but none of the little creatures was able to compete it by then. The second time I set the mutation rate at 100% but only let it run for about 40 generations before I got frustrated and terminated the whole race.   

I'm letting it run for the third time.

[xSilverPhinx]

So...I let the third attempt run for 78 generations, they couldn't get past the slope either. There doesn't seem to be an option to tweak the amount of luck involved.

I changed the gravity to 1.6 (Moon). They bounce and glide a lot but they get further.

I also get this urge to listen to Carmina Burana as I watch them go.   

[Recusant]

Wow, I guess that I was really lucky to see a successful variant in only 50 generations with the default settings. It was a compact one with relatively small wheels that managed to bounce down the terrible terrain at the end of the course.

One of the things I enjoyed was in about the last third of the iterations, there would be a pretty sizable herd of them trundling along through the easier terrain at the beginning.

[xSilverPhinx]

Mine are getting about three quarters through, and end up looking like kites with wheels. 

I like the beginning. When they come into existence there's so much hope and potential.  It doesn't last though. 


 

[joeactor]

Cool - thanks Rec!

[Siz]

Yeah, I'm enjoying watching these go. I'm 70 generations in (letting it run while I'm at work) with default settings. They were stuck on a slope for 20 or so generations but now have advanced to the next unlikely looking mountain. They mostly have largish wheels with a long wheelbase and can run either way up. I'm guessing the terrain is random at the outset?

[Recusant]

I'm guessing the terrain is random at the outset?

The terrain details are random for each session in the default settings, but as it says on the page, "The terrain gets more complex with distance."

[Siz]

My intrepid vehicles were over 120 generations in before some fucker in my office decided to restart my PC while I was out. They were so close to breaking the last mountain too. I wanted to see if/how long it would take. Or, I guess that would necessarily be a 'when'...! When I left them they had shortened their wheelbase and mostly bounced along in somersaults. Is that how they ended up for you?

[Recusant]

My intrepid vehicles were over 120 generations in before some fucker in my office decided to restart my PC while I was out. They were so close to breaking the last mountain too. I wanted to see if/how long it would take. Or, I guess that would necessarily be a 'when'...! When I left them they had shortened their wheelbase and mostly bounced along in somersaults. Is that how they ended up for you?

Yes, the successful model had a small (more or less trapezoidal) medium density body, small, dense wheels, and a very short wheelbase. Its traverse of the bad terrain toward the end of the course was accomplished almost entirely by bouncing.

[xSilverPhinx]

Ok, trial number 6. This time I'm sticking to default settings.

I'm guessing the terrain is random at the outset?

Yes. Each time I tried it was a different terrain. You can also create a world with a seed number. I put '666' this time. 

[Crow]

I have yet to get one to complete the course but I have managed to get an almost identical field of 19 cars that could all get to the same point. That was by doing a rate and size of 2%

Edit: Did it again but with 5%, by gen 30 I had two distinct groups that all got to the same point every now and again you would get one that went a bit further but would fail on a similar section further on. At the higher variables there is too much mutation and not enough passed on to make the heard successful, with the low vales it creates an equilibrium across the cars but there isn't enough variety to get past a problem. I bumped it to 10% on both values you got more variation but not enough passed on to the next generation. I think the answer lies under 10% with an imbalance between the size and rate I'm thinking a lower rate of mutation with a higher size of mutation, thus you get more time for the successful genomes to get passed on but when mutation happens it is enough to make a noticeable difference but not too much you lose the benefits of the original.

[Recusant]

Cool - thanks Rec!

You're welcome.