ASABE Technical Library - Abstract

Click on “Download PDF” for the PDF version or on the title for the HTML version.
If you are not an ASABE member or if your employer has not arranged for access to the full-text, view access and subscription options

3D Perception Framework for Vision-Guided Robotic Harvesting of Button Mushrooms

Published by the American Society of Agricultural and Biological Engineers, St. Joseph, Michigan www.asabe.org

Citation:  2026 ASABE Annual International Meeting  2600142.(doi:10.13031/aim.202600142)
Authors:   Sadjad Mahnan, Mr., Lawrence Arthur, Mr., Long He, Dr., Shirin Ghatrehsamani, Dr., Paul Heinemann, Dr., David Beyer, Dr.
Keywords:   Decision-making, noise removal, point cloud data extraction, principal curvature, spatial configuration.

Abstract. Button mushrooms (Agaricus bisporus) are primarily harvested manually to preserve their quality and shelf life for the fresh market, owing to their delicate skin. However, increasing labor shortages and intensive harvesting practices present considerable challenges to mushroom production and elevate operational expenses. This study presents a solution through the development and evaluation of a selective robotic harvesting system for button mushrooms that mimics human picking actions, including bending, twisting, and lifting. The system incorporates a vacuum-cup-based end-effector, a 3D perception pipeline, and a picking strategy framework to facilitate selective harvesting within growing-bed conditions. A YOLOv8m instance segmentation model was employed to detect individual mushroom caps, complemented by point cloud extraction from RGB-D images. For each detected mushroom, principal component analysis, a 95th-percentile estimator, and principal surface curvature metrics were employed to establish a local reference frame, estimate cap diameter, and evaluate surface geometry for maturity classification. These features were utilized to analyze spatial configurations, facilitating the identification of mushroom vicinity arrangements and surrounding open-space regions. This information determined viable bending directions and picking sequences, which were executed by an end-effector through lateral bending, twisting, and lifting. Experimental results demonstrated a 98.7% mask-based mAP for segmentation, 98.7±0.8% point cloud depth accuracy, and 91.2±5.8% cap diameter estimation accuracy. Maturity assessment achieved a 91.4% F1-score. Open-space region identification achieved a 97.6±2.6% precision, with valid bending direction and picking sequence determination for 97.1% of mature mushrooms.

(Download PDF)    (Export to EndNotes)