MOSU: Autonomous Long-range Robot Navigation with Multi-modal Scene Understanding