Delay-aware model-based reinforcement learning for continuous control

Action delays degrade the performance of reinforcement learning in many real-world systems. This paper proposes a formal definition delay-aware Markov Decision Process and proves it can be transformed into standard MDP with augmented states using reward process. We develop model-based framework that incorporate multi-step delay learned system models without effort. Experiments Gym MuJoCo platforms show proposed algorithm is more efficient training transferable between systems various durations compared state-of-the-art model-free methods.