A Self-Learning Robotic System Solution for Industrial Applications Using Imitation Learning

Artificial Intelligence, robotics, and industrial automation are advancing rapidly to enhance human capabilities and manufacturing efficiency. The UK industry requires self-learning robots that utilise machine and deep learning techniques like Computer Vision and Imitation Learning, which allows robots to learn tasks through observation. This leads to more flexible manufacturing processes with reduced time, complexity, and human error. However, gaps remain between current robotic systems and industry needs. This PhD research aims to develop a self-learning collaborative robotic platform for industrial applications using imitation learning, improving the flexibility, precision, and efficiency of manufacturing robotic cells.
The research focuses on developing a virtual model of a self-learning robotic system for industrial application using imitation learning, leveraging MATLAB/Simulink tools to navigate environments, interact with objects, and make informed decisions. This is achieved through a waypoint tracking algorithm and object detection via convolutional neural networks, enhancing the SLRS's functionality. The SLRS training achieved an average accuracy of 89.02% using SMOTE data balancing techniques, demonstrating viability for real-world applications. We employed the Mycobot 280 Jetson Nano manipulator in SCRIL with a stereo vision algorithm and YOLOv8 for real-time object detection, localising objects in 3D space with K-means clustering. The ROS MoveIt framework facilitated motion planning and control, while real-time object tracking feedback improved the robot's adaptability and responsiveness. By processing video input of human activities involving object handling, the robot can mimic human actions for collaborative assistance, utilizing stochastic gradient descent and Sequential Quadratic Programming for high accuracy and efficiency. Case studies on warehouse and assistant robot applications emphasize the importance of validation and verification in implementing robotic systems, ensuring functionality, reliability, and safety while fostering trust in autonomous capabilities. As robotics technology evolves, ongoing validation will drive advancements and promote acceptance of autonomous technologies in real-world applications.