Using RUBI to partition agents in air traffic problems with hard constraints and reward shaping