Modified SLAM for Agricultural Robot Navigation

The study proposes a novel method for reconstructing missing regions in depth maps, aimed at improving the accuracy of autonomous navigation in agricultural robotic systems. ( Research purpose ) The primary objective is to develop a method capable of compensating for data loss in depth maps, thereby improving the performance of Simultaneous Localization and Mapping (SLAM) systems. ( Materials and methods ) The depth map reconstruction method consists of three main stages: computation of the anisotropic gradient; identification of similar blocks based on a novel similarity criterion; and merging of the detected blocks using a neural network architecture composed of an encoder, a fusion layer, and a decoder. The method was tested using the Rosario dataset, which includes scenarios representative of complex agricultural environments. ( Results and discussion ) The proposed depth map reconstruction method demonstrates a significant improvement in quality metrics. Specifically, Root Mean Square Error (RMSE), Peak Signal-to-Noise Ratio (PSNR), and Structural Similarity Index Measure (SSIM) improved by 20–30 percent compared to the existing techniques. The method preserves the structure and texture features of the reconstructed regions, enabling accurate restoration of large areas with missing pixel data. To evaluate the impact on SLAM performance, the Stereo Multi-State Constraint Kalman Filter (S-MSCKF) algorithm was employed. Quantitative analysis of Absolute Trajectory Error (ATE) and mean RMSE was conducted both before and after applying the depth map reconstruction. The results show a reduction in Absolute Trajectory Error from 0.62 meters to 0.25 meters, and a decrease in Root Mean Square Error from 0.85 meters to 0.39 meters. ( Conclusions ) The proposed method substantially enhances SLAM system accuracy, particularly in challenging agricultural environments, characterized by uneven terrain, variable lighting conditions, and long-distance navigation. Its robust performance suggests strong potential for large-scale integration into autonomous agricultural machinery, contributing to improved reliability, operational efficiency, and safety in robotic field operations.