Model-based Reinforcement Learning with Non-linear Expectation Models and Stochastic Environments