Star Lightweight Convolution and NDT-RRT: An Integrated Path Planning Method for Walnut Harvesting Robots.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Fallen walnut picking robots face challenges with slow response and low accuracy in complex orchards.
• Existing methods struggle to meet the demands of efficient and precise robotic harvesting.

Purpose of the Study:

• To develop an integrated detection and path planning method for fallen walnut picking robots.
• To enhance the speed and accuracy of robotic operations in challenging orchard environments.

Main Methods:

• Proposed YOLO-FW detection model with star-shaped convolution, C3K2 module, CA_HSFPN, and PIoU loss for improved feature extraction and fusion.
• Implemented NDT-RRT path planning algorithm with node rejection, dynamic step-size, and target-bias sampling for efficient planning.
• Developed a detection and planning system on NVIDIA Jetson Xavier NX using PySide6 framework.

Main Results:

• YOLO-FW achieved 90.6% precision, 90.4% recall, and 95.7% mAP@0.5, with a 3.62 MB volume and 30.65% fewer parameters.
• NDT-RRT reduced search time by 87.71% while maintaining path quality.
• On-site tests confirmed the system's robustness, high precision, and real-time performance.

Conclusions:

• The integrated YOLO-FW and NDT-RRT method significantly improves fallen walnut picking robot performance.
• The developed system offers a viable technological solution for intelligent robotic harvesting in real-world orchard conditions.