An Advisor-Based Architecture for a Sample-Efficient Training of Autonomous Navigation Agents with Reinforcement Learning.