I changed the following things to get it to work: ・Made the gap of obstacles a bit larger ・Made the range of biases from -1~1 to -10~10 ・Made the randomness converge slower I still seems to be unstable tho, probably because it lacks cross-over and other important techniques.
