Multi-objective Model-based Policy Search for Data-efficient Learning with Sparse Rewards