Reinforcement Learning in a Physics-Inspired Semi-Markov Environment