Autonomous navigation of Unmanned Ground Vehicles (UGVs) in unstructured environments—where accurate reference paths and well-defined traversable areas are often unavailable—remains a significant challenge. This paper proposes a terrain-aware motion planning framework, TAMP, which achieves robust navigation in unknown off-road environments by integrating coarse satellite map priors with online terrain perception. The system conducts traversability analysis using dense elevation maps generated from LiDAR data and refines the prior path extracted from satellite maps based on real-time environmental constraints. Local paths are generated and filtered through a hierarchical sampling strategy and a terrain-aware path simulation module. Additionally, a pose-aware speed planning strategy is introduced to enhance driving safety in rugged terrain. Extensive real-world off-road experiments validate the effectiveness of the proposed method. With minimal reliance on prior knowledge, TAMP demonstrates strong adaptability to complex environments and offers a practical solution for autonomous driving beyond structured roads.