Session: 03-04-03: Artificial Intelligent Applications in Manufacturing III

Paper Number: 166190

Intelligent Tomato Harvesting: Integrating Industry 4.0/5.0 to Advance Smart Farming Automation 

The integration of Industry 4.0 and Industry 5.0 technologies into Controlled Environment Agriculture (CEA) represents a critical advancement in precision agriculture, smart farming, and autonomous food production. CEA provides a structured and controlled environment ideal for testing and deploying advanced automation, machine learning, and sensor-integrated robotics to optimize yield consistency, labor efficiency, and sustainability. This study presents a Digital Twin-powered robotic harvesting system, specifically developed for high-density indoor farming and automated tomato production, leveraging real-time sensor networks, AI-driven vision, and Industrial Internet of Things (IIoT) connectivity to improve harvesting accuracy and efficiency.

A fundamental problem in modern agriculture is the increasing need for automation to address labor shortages, reduce food waste, and enhance production efficiency in controlled environments. Traditional agricultural automation methods struggle with environmental variability, limiting their applicability in open-field farming. According to the literature review, existing research has identified key gaps in robotic perception, sensor fusion, and adaptive machine learning for CEA applications. While robotic harvesting systems have been explored, many lack the real-time decision-making and predictive analytics capabilities required for scalable deployment. This research seeks to bridge this gap by developing a fully integrated AI-driven robotic harvesting system that leverages Digital Twin technology for simulation and optimization.

This study focuses on the implementation of deep learning-based vision systems and sensor fusion for robotic harvesting. A Digital Twin model was developed using Siemens Tecnomatix software to simulate and validate the robotic harvester’s performance before physical deployment, allowing for iterative improvements in object recognition, robotic path planning, and grasp optimization. The robotic vision system utilizes YOLOv5 and YOLOv11 architectures, trained on a dataset of 348 labeled tomato images, to enable real-time crop classification and selective harvesting. The models were evaluated based on mean average precision (mAP), false detection rates, and computational efficiency, with results showing that YOLOv11 achieved an 88.7% mAP, outperforming YOLOv5 in classification accuracy and adaptability under varying environmental conditions.

In addition to AI-powered vision, sensor-integrated robotic grippers were employed to assess ripeness indices, fruit firmness, and structural integrity, ensuring that only optimally ripe produce was harvested. A predictive analytics framework, leveraging real-time IIoT data streams and cloud-based computation, was implemented to optimize harvest scheduling, minimize post-harvest losses, and enhance supply chain efficiency. Comparative trials demonstrated that AI-enhanced decision-making improved harvesting efficiency by 26% over traditional automation techniques, with reductions in crop damage, operational costs, and energy consumption. The Digital Twin simulations allowed for further parameter optimization, identifying key variables that contributed to robotic harvesting precision and reliability.

CEA provides an ideal testbed for scaling autonomous robotic solutions due to its high repeatability and controlled environmental factors, differentiating it from open-field agriculture, where variability poses significant challenges for automation. The ability to integrate real-time AI decision-making, robotic manipulation, and precision sensor technology aligns CEA with the future of advanced agricultural automation. This research addresses key challenges in robotic perception, intelligent actuation, and predictive modeling, establishing a scalable and data-driven framework for commercial smart farming operations.

This study contributes directly to agricultural extension efforts, bridging the gap between engineering advancements and real-world farm adoption. As labor shortages and sustainability concerns drive the need for next-generation automation in agriculture, this research provides a replicable and adaptable model for integrating Digital Twin-based robotics into vertical farming and other controlled-environment systems. Future work will focus on multi-crop adaptability, enhanced AI-driven yield forecasting, and the integration of cloud-based farm management platforms to further advance scalable and autonomous food production solutions.

Presenting Author: Jacob Holloway Kennesaw State University

Presenting Author Biography: Jacob Holloway is an agricultural scientist specializing in agricultural robotics and artificial intelligence (AI) with applications in Controlled Environment Agriculture (CEA). With a background in agricultural extension, he has worked on integrating advanced automation, digital twin technology, and AI-driven decision-making into precision farming systems. His research focuses on developing intelligent sensor networks for real-time monitoring in greenhouse environments, optimizing climate control, and improving farm productivity through robotics and data-driven solutions. Jacob’s work leverages Tecnomatix simulations, hyperspectral imaging, and Industrial Internet of Things (IIoT) technologies to advance sustainable and scalable indoor farming practices.

Authors:

Jacob Holloway Kennesaw State University
Denzel Oden Kennesaw State University
David Guerra-Zubiaga Kennesaw State University
Gershom Richards Kennesaw State University