Abstract: This research addresses the significant challenges of robot localization and navigation within the complex and dynamic environments characteristic of modern sheep farms. Factors such as moving animals, changing obstacles, and variable terrain contribute to positioning difficulties and reduced navigational accuracy for inspection robots. To overcome these issues, a comprehensive high-precision 3D laser-based positioning and autonomous navigation system is developed, specifically engineered for reliable operation in real sheep farm conditions.The system architecture begins with robust environmental perception. A three-dimensional Light Detection and Ranging (LiDAR) sensor is tightly coupled with an Inertial Measurement Unit (IMU). This sensor fusion scheme provides rich spatial data and inertial motion cues. A tightly coupled LiDAR-inertial odometry and mapping algorithm is employed to construct a consistent and accurate navigation map of the environment, forming the fundamental spatial reference for all subsequent operations.A core innovation of this work lies in its multi-stage approach to handling dynamic interference, primarily caused by moving sheep. Initially, a viewpoint visibility method is applied to perform preliminary filtering of the acquired point cloud data, identifying and removing obvious transient points. Building upon this and incorporating concepts from the Egocentric Ratio of pSeudo Occupancy-based Dynamic Object Removal (ERASOR) paradigm, an enhanced dynamic point detection algorithm is proposed. This method synergistically leverages both height and distance features from the point cloud to more intelligently distinguish between static structural elements and dynamic obstacles. This two-tiered filtering strategy significantly purifies the point cloud data, substantially reducing the corrupting influence of dynamic objects on the mapping and localization processes.Precise and stable robot positioning is achieved through a hierarchical localization strategy. For accurate local pose estimation, an Error State Extended Kalman Filter (ESEKF) fuses high-frequency data from the laser odometry (derived from the filtered point clouds) and the IMU. This ensures smooth and precise tracking of the robot's immediate movements. For robust global localization, particularly to mitigate drift or recover from potential tracking failures, an enhanced adaptive Monte Carlo localization method is implemented. This global estimator is strengthened by integrating the Normal Distribution Transform (NDT) with the Iterative Closest Point (ICP) algorithm, improving its accuracy and convergence in feature-varying agricultural scenes.Navigation is facilitated by a hybrid path planning methodology. The global path is initially computed using the A* search algorithm on the built map, providing an optimal long-range route. This global path is then optimized in real-time by the Timed Elastic Band (TEB) local planner. The TEB algorithm dynamically adjusts the robot's trajectory, considering temporal constraints, kinematic model, and immediate sensor observations to generate smooth, collision-free, and executable velocity commands.Experimental validation in an operational sheep farm environment confirms the system's efficacy. Compared to conventional Simultaneous Localization and Mapping (SLAM) approaches without specialized dynamic filtering, the proposed dynamic point cloud removal algorithm markedly enhances localization precision. Quantitative results show improvement rates of 35.2% in average lateral deviation and 28.7% in longitudinal deviation, leading to an overall positioning accuracy increase of 31.8%. During field operations at speeds between 0.3 and 0.5 meters per second, the system maintained an average heading deviation below 2.4 degrees (standard deviation < 3.2°). Both lateral and longitudinal positional deviations exhibited average values under 3.5 centimeters, with standard deviations below 2.9 centimeters. Performance was evaluated across three motion modes: forward, backward, and line-changing. While forward motion demonstrated the highest accuracy, the slightly reduced precision observed in backward and line-changing modes remained well within the operational requirements for autonomous agricultural robotics.In conclusion, this study presents a viable and effective 3D laser positioning and navigation solution that successfully counters the complexities imposed by dynamic farm environments. By achieving high-precision mapping, resilient localization, and adaptable navigation, it ensures reliable autonomous mobility for robotic platforms in sheep farms. This work provides a substantive technical foundation for the wider deployment of autonomous mobile systems in challenging and unstructured agricultural settings.