This is way more fun than seems reasonable. This program uses a simple genetic algorithm to evolve random two-wheeled shapes into cars over generations. From what I can see it is based purely on random mutation* of parameters where the one with the best performing "phenotype" is preserved and copied to the next generation.
Anyway, try running it in a background tab for a few hours.http://rednuht.org/genetic_cars_2/
* Some genetic algorithms go further than this by also "cross breeding" the best performers (ie by making a hybrid parameter set using bits of each parent). Presumably, you can get better designs faster this way because you can explore the solution space without necessarily getting trapped in nearby but sub-optimal solutions.