Reward Function for q learning on a robot

I have 2 wheeled differential drive robot which I use pid for low level control to follow line. I implemented q learning which uses samples for 16 iterations then uses them to decide the best position to be on the line so car takes the turn from there. This allows PID to setup and smooth fast following. My question is how can I setup a reward function that improves the performance i.e. lets the q learning to find the best What it tries to learn is this, it has 16 inputs which contains the line positions for the last 15 iterations and this iteration. Line position is between -1 and 1 which -1 means only left most sensor sees the line and 0 means the line is in the center.