Abstract

This dataset provides the source code and trained model weights used to implement and evaluate a graph-based approach for the perception and topological representation of Deformable Multi-Linear Objects (DMLOs), also referred to as Branched Deformable Linear Objects (BDLOs). These objects, such as automotive wiring harnesses, exhibit complex branching structures that are challenging for robotic perception and manipulation. The released material enables the reproduction of a pipeline that estimates a coherent graph representation of DMLOs starting from binary scene masks. Graph nodes are sampled along estimated object centerlines, while a graph neural network is employed to predict graph connectivity and to classify node types based on local topology and orientation. A solver then combines these predictions to generate a consistent topological representation of the objects. The code and model weights were produced in the framework of the Horizon Europe IntelliMan project and were used in the experiments presented in the associated publication: A. Caporali, K. Galassi, R. Zanella and G. Palli, "GNN Topology Representation Learning for Deformable Multi-Linear Objects Dual-Arm Robotic Manipulation", in IEEE Transactions on Automation Science and Engineering, vol. 22, pp. 14738-14751, 2025, DOI: 10.1109/TASE.2025.3562231.