You can try to rework their decisions, the source code is right there. You could try to make a neural network that uses the three inputs and maybe two in 45 degree angles to output a value for each option(left, right, straight)

I have tried neural networks once, but failed;-) The ai uses raycasts; it checks the distance in every possible direction (left, right, straight) and selects the direction with the biggest distance. With the difficulty slider you basically adjust the frequency at which the ai checks the distances. Also, the update frequency is randomized