Operationally Responsive Textile Manufacturing using Collaborative Robots

The U.S. defense industrial supply chain currently lacks highly automated textile manufacturing capabilities.  Manual loading of sewing machines and textile processes can account for up to 40% of labor costs. Production volumes in the textile manufacturing industry are further challenged by workforce fluctuations and shortages of skilled workers. This project aims to develop an automated sewing robot system that leverages AI, computer vision, and robotics to enhance textile manufacturing for defense and commercial applications. In Phase I, the team will develop a deep learning-based image segmentation model specifically designed to handle various textiles and industrial materials, including camouflage, nets, and thermal blankets. The project will involve creating a custom labeled dataset and employing data augmentation techniques to increase the model’s robustness. Utilizing the PyTorch framework, the team will explore state-of-the-art architectures such as U-Net and conduct training on cloud platforms with GPU and TPU resources, focusing on minimizing latency between fabric manipulation and model feedback. Upon achieving target accuracy, the model will be integrated with Standard Bots RO1 robotic arms using Robot Operating System (ROS) 2 and NVIDIA Isaac ROS, enabling autonomous tasks such as fabric alignment, flattening, and layering. Integration efforts will include developing a control codebase in Python and C++, and creating a graphical user interface (GUI) for intuitive human interaction. Phase I will result in a functional prototype system, evaluated on multiple fabric manipulation tasks, with an assessment of integrating sewing capabilities and identifying areas for further development.