Bayesian inference for data-efficient, explainable, and safe robotic motion planning: A review