Learning-based Hamilton-Jacobi-Bellman Methods for Optimal Control.