Improving the K-nearest neighbor algorithm with a Genetic algorithm

I am using the Aproximate Nearest Neighbor library on a trajectory following problem. My input values are speed, distance to next point, angle to next point etc., and my output should be gas and break pedal value as well as the steering angle. I've been getting pretty bad results, so I decided to write a simple Genetic algorithm and try to find the best parameters for the KNN algorithm. How can I make the most of this combination? Would the weighting of the input values be a good idea?

I'm fairly new to machine learning, so any advice would be greatly appreciated