The paper presents UniFolding, a scalable, sample-efficient, and generalizable robotic system for folding different types of garments. The system uses the proposed UFONet neural network to combine unfolding and folding decisions into one adaptable policy model. The design of UniFolding is based on a garment’s partial point cloud, which helps with generalization and reduces sensitivity to variations in texture and shape. Training data is collected through a human-centric process with offline and online stages. The offline stage involves human unfolding and folding actions via Virtual Reality, while the online stage uses human-in-the-loop learning to fine-tune the model in a real-world setting.
Publication date: 2 Nov 2023
Project Page: https://unifolding.robotflow.ai
Paper: https://arxiv.org/pdf/2311.01267
