Learning Control Policies for Variable Objectives from Offline Data

Offline reinforcement learning provides a viable approach to obtain advanced control strategies for dynamical systems, in particular when direct interaction with the environment is not available.