Digital Twin System Framework for Gripping Deformable Objects with Synthetic Data and Reinforcement Learning

This paper details the development of an advanced robotic system utilizing digital twin technology for the efficient manipulation of deformable objects, such as cables and tubes, within manufacturing processes. Traditional robotic systems often require human intervention for these tasks, leading to increased operational costs and delays. Our system dynamically updates its programming to handle these materials effectively, using synthetic data generated from a virtual environment to improve both efficiency and adaptability in manufacturing. The approach integrates image-based detection and reinforcement learning for accurate manipulation, alongside a dynamic feedback mechanism for addressing complex handling tasks using offline programming. Furthermore, the system employs a probabilistic roadmap algorithm for generating robot path without obstacle avoidance. The paper outlines the research problem, reviews relevant literature, describes the proposed methodology, and presents experimental results. These demonstrate the system’s enhanced capability in accurately identifying and managing flexible objects, representing a significant improvement in robotic technology for the manufacturing industry.