Dynamical systems represented using probabilistic models are hindered by the model bias induced by the prior, which reduces data efficiency. Herein, we explore how to construct physics-informed priors to increase data efficiency and mitigate the sim2real gap.
