Semi-Markov Decision Processes (SMDPs): a formalism for describing Reinforcement Learning problems that involve temporally extended decisions or actions. Such actions are often implemented as controllers that encapsulate a class of useful behavior. See control basis.