I presented our short-paper at the ML4CPS conference. In the paper we discuss the potential of leveraging self-supervised pretraining for reinforcement learning from human feedback in the context of cyber-physical systems.