Hilly orchards are characterized by undulating terrain, canopy occlusion, irregular tree-row structures, and marked variations in ground adhesion conditions, which challenge the autonomous navigation of agricultural machinery by degrading perception and localization, increasing reference path uncertainty, and reducing closed-loop robustness. Perception and localization errors can propagate progressively along the chain of “sensor observation–vehicle pose estimation/environmental-structure perception–reference path generation–path-tracking controller inputs–vehicle closed-loop response,” ultimately affecting path-tracking accuracy, control smoothness, and operational safety. This review examines perception and localization error propagation and robust path-tracking control for agricultural machinery in hilly orchards. The review first summarizes the navigation roles and error characteristics of key observation sources and then analyzes how perception, localization, and reference path generation provide state variables, reference variables, and safety constraints to controllers. It subsequently compares typical path-tracking methods under multi-source disturbances and actuator constraints and discusses key challenges and an integrated robust design direction coupling perception and localization, reference path generation, and path-tracking control. The proposed integrated framework represents a synthesis-derived design direction rather than a complete architecture that has already been experimentally validated in hilly orchard field environments.

Navigating hilly orchards is challenging for autonomous agricultural vehicles due to the rugged terrain and dense canopy cover. Standard environmental modeling techniques are widely used, yet they often overlook how elevation uncertainty propagates during Digital Elevation Model (DEM) reconstruction. This oversight can directly affect terrain risk assessments and navigation planning. From an error-propagation perspective, this review examines how uncertainties originating from RTK-GNSS, LiDAR, and computer vision propagate through DEM reconstruction, terrain-feature extraction, cost map construction, and path planning. We further analyze how DEM elevation errors and vertical inaccuracies affect slope estimation, roughness representation, traversability assessment, vehicle stability, and navigation safety. Finally, we highlight practical bottlenecks in hilly orchard scenarios and suggest several research priorities, including multimodal fusion, uncertainty-aware modeling, lifelong map updating, and learning-based traversability assessment.