Reinforcement Learning in Partially Observable Markov Decision Processes using Hybrid Probabilistic Logic Programs