Learning State Switching for Improved Exploration in Multi-sensor Environments