Multi-source Information Fusion-based Autonomous Navigation System for Vehicles in Swine House

Abstract:

This paper presents an autonomous navigation system for unmanned pig farming, using multi-source information fusion. It integrates LiDAR and IMU data via the Fast-LIO2 algorithm for odometry, constructs maps with a factor graph-optimized loop closure method, and localizes using both map registration and reflective pole matching, outperforming Adaptive Monte Carlo Localization. Path planning is handled by Dijkstra's algorithm for global paths and the Time-Elastic Band for local paths, ultimately realizing autonomous navigation in pig house scenarios. Experimental results show that the system achieves mapping with a maximum absolute error of 0.077 m and relative error of 3.79%, with higher accuracy compared to mapping algorithms without loop closure detection. Localization accuracy averages 0.066 m in X and 0.052 m in Y for map registration, and 0.046 m in X and 0.042 m in Y for reflective pole matching. When the mobile platform navigates at a speed of 0.3 m/s, the maximum lateral deviation between the actual navigation points and the target points is 0.09 m, the maximum longitudinal deviation is 0.089 m, and the average heading angle deviation is 7.06°. The system meets high-precision requirements for mapping, localization, and navigation in swine houses, supporting unmanned pig farming operations.