AutoNeRF: Training Implicit Scene Representations with Autonomous Agents