Learning Inverse Statics Models Efficiently with Symmetry-Based Exploration