Determining the follow-wall reference vectors turns out to be a bit more involved than either the avoid-obstacle or go-to-goal reference vectors. Take a look at the Python code in follow_wall ...