>>7615419You simply program it to to use non-linear pathways based on trial and error.
For example if you wanna make a robot walk, you make it do things and move the limbs and compare it to the corresponding demanded result. If whatever the robot does is close to the result, it keeps that action in memory and gives it a percentage value. If whatever the robot does is not useful it gives it a low percentage value. After a number of mutations, it can select the useful ones, like say the top %70 and start mutating those actions to see if it gets any closer to walking.
By this method, the robot will find the most efficient way to use its limbs correctly and walk by itself.