Humans acquire their most basic physical concepts early in development, and continue to enrich and expand their intuitive physics throughout life as they are exposed to more and varied dynamical environments. This project introduces a hierarchical Bayesian framework to explain how people can learn physical theories from observation across multiple timescales and levels of abstraction. We work with expressive probabilistic program representations suitable for learning the forces and properties that govern how objects interact in dynamic scenes unfolding over time. The project compares the model with human learners on a challenging task of inferring novel physics in microworlds given short movies. This task mimics the real developmental challenge of learning physics more closely than prior experimental work, by requiring people to reason simultaneously about multiple interacting physical laws and properties. In addition, the project explores rational approximations to an ideal-level model, which complements a top-down Bayesian approach with a more bottom-up feature-based approximate inference scheme.