Addressing Agricultural Robotic (Agribots) Functionalities and Automation in Agriculture Practices: What’s Next?

Abstract

This paper presents and explores the functionalities of automation and agricultural robotics (agribots) in recent years for agricultural operations. The explicit challenges fronting agribots and automation with regards to the operative implementation of Industry 4.0 are discussed. In this paper, several research works and developments on automation and agribots from different scopes and field areas are reviewed to explore recent agricultural practices. The first technology is on the automation work on developing a control algorithm that uses a single sensor that could recognise landmarks in the row-type plantation environment. This is followed by the navigation of a vehicle using a laser range finder (LRF) to a point-to-go aim location in the plantation by generating a control algorithm equipped with a sensor for an autonomous agricultural vehicle to detect the landmarks in the row-type plantation setting. The second technology is related to an automation device by developing an automated detector and counter (Oto-BaC^TM) for bagworm census using deep learning with a Faster Regional Convolutional Neural Network (Faster R-CNN) and normal camera. Meanwhile the third technology is related to agribots via the design and development of a fully automated Agriculture robot (Agribot) for harvesting underground plants (rhizomes) with the assistance of transmission and receiving parts using microcontroller software. Another agribot technology would be on the development of Thorvald II agricultural robotic system, utilising a modularity hardware whereby the robot consists of standard modules and can be reconstructed to handle tasks in various types of environments. The first automation technology results showed the performance of the navigation systems to operate the tractor autonomously along the test path without any crashes on the guide cones. The second automation technology on the Oto-BaC^TM performance, produced a positive Pearson product-moment correlation coefficient between the two variables (percentages of detection and temperature), R² = 0.997 and p = 0.02 for Trial 1 and R² = 0.888 and p = 0.04 for Trial 2. Meanwhile, the third technology on the Agribot, successfully picked up the rhizome plants, sprayed pesticides, and traced of the soil moisture content. Finally, the last automation technology which was on the development of Thorvald II, came out with positive results on the pass traction test, all

obstacles, and the incline test. The harvesting robot detected the ripe tomatoes at a 95% success rate by implementing the self-developed algorithm that applied the Adaboost and APV classifiers. However, a 5% miss detection occurred due to the leaf obstruction. The multi-robot system can be designed to handle pest control tasks via UAVs and UGVs. For weed patch recognition, the developed algorithms showed their robustness by precisely distinguishing and mapping the crop rows with a 100% accuracy, while the inter-row weed patches with an accuracy of 85%, and it was proposed to detect the early growth stage based on the weed maps through site-specific weed management. By implementing a 3D fruit detection algorithm, the precision for pepper, eggplant, and guava datasets was 0.864, 0.886, and 0.888, while the recall dataset was 0.889, 0.762, and 0.812, respectively. The proposed algorithm was effective and robust; hence it is appropriate to be applied as an agricultural harvesting robot. A roadmap in applying swarm robotics is described towards the weed control problems and is being implemented within the Swarm Robotics for Agricultural Applications scope. Thus, a baseline result was introduced specifically for monitoring and mapping out weeds in a field via a swarm of UAVs. Hence, the impact on the use of agribots and automation technologies is the realization of a more efficient systems potential to be operated in safe conditions and are cost effective for the farmers, allowing farmers to focus more on improving overall production yields. As a recommendation, there is a need for research and development of multipurpose and adaptive algorithms to be incorporated into different sensor platforms.