Offline reinforcement learning: Tutorial, review, and perspectives on open problems. arXiv preprint arXiv:2005.01643, 2020. Google Scholar Timothy P Lillicrap, Jonathan J Hunt, Alexander Pritzel, Nicolas Heess, Tom Erez, Yuval Tassa, David Silver, and Daan Wierstra. Continuous control with deep reinforcement learning. However, offline or data-driven RL overcomes these challenges by removing the agent's dependence on interacting with the environment and instead using historical data for training. This study proposes to train an agent with synthetic combat simulation data using offline or data-driven RL. Download the PDF conference document. Offline reinforcement learning should develop the policy better than the given dataset, which means that the learning action should be different from the corresponding one in the dataset. But current machine learning methods rely on the assumption that the data is independent and identically distributed, and pursues the underlying distribution of the. An offline model-based RL method learns an ensemble model through supervised learning and incorporating uncertainty penalties into the feedback. COMBO, 18 A conservative offline model-based policy optimization method regularizes the value functions of unsupported state-action tuples generated under the learned model.