Continuous-Time Model-Based Reinforcement Learning

                agatay Yldz 1 Markus Heinonen 1 Harri Lhdesmki 1

             Abstract
    Model-based reinforcement learning (MBRL) ap-
    proaches rely on discrete-time state transition
    models whereas physical systems and the vast ma-
    jority of control tasks operate in continuous-time.
    To avoid time-discretization approximation of the
    underlying process, we propose a continuous-time
    MBRL framework based on a novel  ...