DESIGN OF A CONTROL SYSTEM FOR THE MOVEMENT OF A SWARM OF SPHERICAL ROBOTS WITH A DIGITAL TWIN APPROACH AND DECENTRALIZED MULTI-AGENT LEARNING USING Q-LEARNING ALGORITHMS

• Main Author: Hubert, Moses
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/69953
• Institution: Institut Teknologi Bandung

In the field of robotics, the development of a multi-agent system using several robots is called a multi-robot system. This system was developed to solve various complex problems that cannot be solved by a single robot. This research focuses on designing control systems for swarm of robots. The design of the control system is designed to resemble a cascade controller design with the primary controller in the form of a Q-learning algorithm and the secondary controller in the form of a PI controller. The Q-learning algorithm is used so that each robot can learn to work together in a decentralized manner and apply the concept of swarm robotics. This concept is used to make multi-robot system more scalable, robust and flexible. In this study, the robot used is a spherical robot because it has holonomic movement. In addition, to overcome the problem of the limited number of robots and the high cost of buying them, this study tested a multi-agent system using a digital twin approach. With this approach, adding the number of agents to the system can be done more easily. This research focuses on four stages: developing a robot detection system, developing a digital robot model, implementing the Q-learning algorithm for agent training, and testing training results with a digital twin approach. Identification of robot position is done by color segmentation method. The digital model is developed by determining the mathematical equation of the speed and orientation of the robot. The training algorithm is used so that agents can make the best decisions on testing.