Autonomous Agricultural Robot Framework featuring Temporal Dynamic Graph Neural Network based Semantic Segmentation coupled with Novel Navigation Line Extraction for efficient field operation and intelligent decision making
Related Experiment Video
Updated: Nov 6, 2026
Closed-loop Neuro-robotic Experiments to Test Computational Properties of Neuronal Networks
Published on: March 2, 2015
Sakthivel Kirubakaran1, Surapaneni Renu Deepti2, Yetcherla Kumar Sekhar3
1CMR College of Engineering and Technology, Hyderabad, India.
Background:
Autonomous agricultural robots support precision farming and improve operational efficiency; however, reliable navigation in complex field environments remains challenging because of inconsistent lighting, variable soil textures, irregular crop growth patterns, and visual occlusions. These conditions hinder accurate crop row detection, semantic segmentation, and navigation line extraction. This study proposes an Autonomous Agricultural Robot Framework Featuring Temporal Dynamic Graph Neural Network Based Semantic Segmentation Coupled with Novel Navigation Line Extraction (AAR-TDGNN-SS-NLE) to improve navigable path identification and autonomous field operation.
Results:
RGB images of corn crop rows from the Crop Row Detection Dataset are annotated with ground-truth masks to distinguish traversable and non-traversable regions. Image preprocessing using the Inverse Unscented Kalman Filter (IUKF) includes noise reduction, normalization, resizing, and contrast enhancement. Geometric transformations and color adjustments are applied to increase dataset diversity and improve model generalization. Efficient Net-B7 performs pixel-level semantic segmentation, while the Human Evolutionary Optimization Algorithm (HEOA) optimizes model parameters. The Temporal Dynamic Graph Neural Network (TDGNN) subsequently extracts stable navigation lines by capturing temporal relationships, path curvature, and orientation. Performance is evaluated using accuracy, precision, recall, F1-score, and intersection over union (IoU). The proposed AAR-TDGNN-SS-NLE framework achieves an overall accuracy of 98.01% and outperforms the existing DLSS-NNLE-AAR, SS-EID-FRNLE, and UDA-SS-AR frameworks according to the reported experimental results.
Conclusion:
The proposed framework integrates semantic segmentation, evolutionary optimization, and temporal graph-based navigation line extraction to support accurate path identification and autonomous agricultural navigation. It offers potential for improving the efficiency of agricultural field operations and supporting intelligent decision-making in complex farming environments. © 2026 Society of Chemical Industry.
