Continuous Control for High-Dimensional State Spaces: An Interactive Learning Approach