DESIGN, INTEGRATION, AND EVALUATION OF A DUAL-ARM ROBOTIC SYSTEM FOR HIGH-THROUGHPUT TISSUE SAMPLING FROM POTATO TUBERS

Divyanth Loganathan Girija

doi:10.7273/000007411

Molecular-based pathogen detection from potato tubers offers a promising alternative to traditional post-harvest grow-out tests, which are field-based, time-consuming, and costly. However, manual tuber tissue extraction for molecular detection remains a significant bottleneck as the process is labor-intensive and physically demanding. To overcome these limitations, this study developed an automated tissue sampling system using a machine-vision-guided dual-arm coordinated inline robotic platform. This system integrated tuber grasping and tissue sampling mechanisms, enabling precise and rapid tissue extraction from specific regions of the tuber, such as the eyes and stolon scars. The first phase of the study investigated various deep-learning-based object detectors, including multiple variants of the You Only Look Once (YOLO), namely, YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLOv9, YOLOv10, and YOLOv11, for detecting eyes and stolon scars from a robust image dataset obtained from tubers of five potato cultivars (three russet-skinned, one red-skinned, and one purple-skinned). The class-wise mean average precision at an intersection over union threshold of 0.5 (mAP@0.5) ranged from 0.832–0.854 with YOLOv5n to 0.903–0.914 with YOLOv10l across the cultivars. Among all the tested models, YOLOv10m demonstrated the optimal balance between detection accuracy (mAP@0.5 of 0.911) and inference time (92 ms), along with satisfactory generalization performance when cross-validated across the cultivars used in this study. The model benchmarking and findings from this phase of the study provided valuable insights into the development of a reliable machine-vision system for robotic tuber sampling applications. In the second phase, an in-line dual-arm robotic system incorporating a machine-vision-based control scheme was developed. The system functioned by transporting tubers onto a conveyor, which halts when a YOLO11-based vision system detects a tuber entering the workspace of the tuber-gripping mechanism. The gripping-mechanism consisted of a one-prismatic (P)-degree-of-freedom (DoF) arm equipped with a gripping end-effector that securely grasped the tuber and positioned it for sampling. The second robotic arm, a 3-P-DoF Cartesian manipulator with a biopsy punch-based end-effector, performed tissue sampling. Tissue sampling was carried out at specific locations on the tubers (eyes or stolon scars), which were detected and localized by the YOLOv10m-enabled vision system. A four-stage tissue-sampling process was introduced, involving the insertion of the end-effector into the tuber, rotation of the biopsy punch for tissue detachment from the tuber, biopsy punch extraction, and deposition of the tissue core onto a designated collection area. The results showed an average positional error of 1.84 mm in reaching the sampling location along the tuber surface, and an average depth deviation of 1.79 mm from the intended 7.00 mm punching depth. The system achieved an overall success rate of 81.5% in tissue core extraction and deposition onto the designated collection area, with an average sampling cycle time of 10.4 seconds. With a total cost of system components less than $1,900, the system demonstrates promising potential as a cost-effective alternative to labor-intensive manual tissue sampling. To summarize, this study developed a robotic platform that offers a cost-effective and scalable solution for potato tuber tissue sampling thereby facilitating the adoption of high-throughput molecular-based diagnostic workflows.

DESIGN, INTEGRATION, AND EVALUATION OF A DUAL-ARM ROBOTIC SYSTEM FOR HIGH-THROUGHPUT TISSUE SAMPLING FROM POTATO TUBERS

Files and links (1)

Abstract

Metrics

Details