Task-level imitation learning using variance-based movement optimization.