Development of a deep learning system for vision based robot control

• Main Author: Ahmed, Rashna Analia
• Other Authors: Cheah Chien Chern
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Deep learning
• Online Access: https://hdl.handle.net/10356/177213
• Institution: Nanyang Technological University

This report presents the development of a deep learning system designed to enable a robot to autonomously perform the task of watering plants. The system utilises a Convolutional Neural Network (CNN) based on the YOLOv5 OBB architecture for rotated object detection, allowing the robot to accurately identify and locate a watering can in its environment. Once detected, the robot is trained to pick up the watering can, rotate it above a pot, and simulate the act of watering. The report details the methodology used in training the CNN, the integration of the deep learning model with the robotic control system, and the results of the system's performance in a controlled environment. The implications of this technology for automating routine agricultural tasks and its potential applications in precision agriculture are also discussed.