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

Reliable Fault Detection in Depth Estimation for Robotic Picking Operations

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

Citation:  2026 ASABE Annual International Meeting  2600776.(doi:10.13031/aim.202600776)
Authors:   Efi Safikou, Yunjun Xu
Keywords:   Fault detection, symbolic regression, inferential sensing, physical AI, estimation theory, robotics

Abstract. Reliability is a key factor that constrains the adoption of robotic tools in agriculture, particularly in small- and medium-scale farms. Such systems are prone to faults during field operations and require frequent maintenance, which makes them costly for farmers. For example, recent strawberry harvesting experiments indicate that a significant portion of harvesting failure cases can be attributed to an inaccurate height estimation. Therefore, the development of an advanced fault detection tool is crucial for guaranteeing the accuracy and reliability of sensed height information. This study introduces a hybrid method that combines physics-based and data-driven methods for creating an informative inferential/soft sensor that achieves robust fault detection in noisy and uncertain environments, thereby providing a cost-effective alternative to conventional depth-sensing hardware. The proposed approach leverages symbolic regression via genetic programming to derive mathematical functions that accurately estimate system parameters, even when data availability is limited. These functions are optimized using the Ds-optimality criterion to enhance fault sensitivity beyond conventional (hardware) sensors. The robustness of the proposed method is assessed using a k-nearest neighbors classification algorithm to compare the fault detection accuracy of hardware versus the inferential sensor and to quantify the improvement achieved through the integration of the inferential sensor. Comparative results show that inferential sensors can improve accuracy and robustness in fault detection, while effectively mitigating the impact of uncertainty and measurement noise.

(Download PDF)    (Export to EndNotes)