Automation in a Two-Wheeled Drive Hand Tractor Using the Embedded Control System

Abstract

The diminishing agricultural land is a problem that is being faced by developing countries like Indonesia. According to the Indonesian Central Statistics Agency, there was a significant reduction in agricultural land from 2014 to 2018. One reason is that the effort spent cultivating or ploughing the land is not worth the income obtained by farmers.This is what makes farmers decide to convert their agricultural land. This will be undeniably a serious problem for Indonesia's food security. Therefore, it is essential to combine conventional agriculture with technology in the form of a tractor that can operate automatically to help farmers ploughing the paddy fields. This dissertation presents the design, manufacture, and implementation of automation on a two-wheeled drive hand tractor. The five objectives of this dissertation were to: 1) design and create software, electrical and mechanical control system of a 2WD hand tractor, 2) collect the sensors data set of the 2WD hand tractor control system, 3) develop the path planning navigation using GIS and HTML, 4) develop and analyze GPS, accelerometer, gyroscope and compass data fusion system using the Kalman filter applied to the 2WD hand tractor, and 5) develop and analyze a Rice Field Sidewalk (RIFIS) detection system using image processing. The methodology used to achieve these objectives included the collection of the tractor's behavior data that were controlled manually and remotely. The path-planning algorithm using the waypoint navigation method was used to complement the autonomous capabilities of the tractor. Several sensors were installed on the tractor (GPS, compass, and camera), and the data were collected using the MQTT Internet of Things protocol. Furthermore, it is necessary to apply a sensor fusion to overcome the noise generated by the sensor during data collection. The method applied both the Kalman filter and the Butterworth filter. Meanwhile, the camera installed on the tractor produced video dataset recordings used as an input for the Rice Field Sidewalk detection process. The Mask-RCNN method was selected and tested as a detection algorithm. In this study, the five primary objectives have been accomplished. The initial objective was to construct TROLLS: Tractor Controlling System, which combines software, electrical, and mechanical components to enable the remote tractor control. The initial prototype tested on the tractor Quick G-3000 was then reviewed. The study and assessment results were then applied to the Quick G-1000 tractor. The second objective was achieved during the field trial by recording the sensor and video data. This Rice Field Sidewalk (RIFIS) dataset is a compilation of GPS, compass, and camera sensor readings utilized to accomplish objectives four and five. In addition, the third objective involves the development of a path-planning platform based on Laravel and Google Maps, with the starting point, ending point, and puddler's distance serving as initial inputs. As an automation strategy for the tractor, the generated path results are deployed as an input for the waypoint navigation. The sensor readings are less steady based on the data acquired for the third objective; the Kalman filter eliminates the noise. For the fifth objective, an early investigation was done by utilizing Deep Learning approaches to detect RIFIS. Based on the research, an autonomous tractor control system has been created to help farmers in the process of ploughing fields.