FIeld self-delivery system using quadruple legged robot and robot manipulator

• Main Author: Koo, Jie Hui
• Other Authors: Xie Lihua
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Electrical and electronic engineering::Control and instrumentation::Robotics
• Online Access: https://hdl.handle.net/10356/167556
• Institution: Nanyang Technological University

This project explores the use of a legged robot dog, the design and fabrication of an accompanying payload, as well as implementation of perception and behaviour algorithms used to pick up and drop off objects in the context of an urban workplace environment, specifically in an urban farm.However, the solutions proposed can also be extended to other workplace environments. Various forms of robot locomotion exist in the robotics industry, including wheeled robots and tank tracked robots. In an urban environment, these may not be as suitable as the urban environment is designed for humans. Therefore, legged robots are increasingly chosen for its ease of mobility and navigation in urban environments. Workplace delivery of items is menial and labour intensive. Automating the delivery of these items will allow workers to focus on the task at hand, reducing the need to leave the work area. This can significantly improve productivity. Data collection and tracking of physical assets is also a key issue in the workplace. Usually manual and labour intensive, workers are needed to physically examine the equipment, and data such as photos taken may not be in the same position and place, making comparisons over a period of time difficult. With the use of robots for photo data collection, reliable and consistent data can be collected while reducing manpower. Several key achievements were attained in this project. Firstly, mechanical design and fabrication of the payload, including the manipulator and mounting of sensors and on-board computers. Secondly, a novel intersecting bounding box approach for locating and identifying objects is proposed and implemented. It uses fused image and point cloud data and needs neither machine learning for locating objects, nor computing intensive point cloud algorithms like template matching. Thirdly, complex behaviours for picking up and dropping off objects are implemented in a behaviour tree, as well as reliably aligning to target locations, showcasing the flexibility and extensibility of the robot behaviour implementation in this project. Fourthly, a data collection pipeline and website are implemented with HTML, PHP and mySQL.