Co-Manipulation of Soft-Materials Estimating Deformation from Depth Images

Human-robot co-manipulation of soft materials, such as fabrics, composites, and sheets paper/cardboard, is a challenging operation that presents several relevant industrial applications. Estimating the deformation state co-manipulated material one main challenges. Viable methods provide indirect measure by calculating human-robot relative distance. In this paper, we develop data-driven model to estimate from depth image through Convolutional Neural Network (CNN). First, define roto-translation current robot pose human grasping position. The estimates Network, specifically DenseNet-121 pretrained on ImageNet.The delta between desired fed controller outputs twist commands. paper describes developed approach acquire, preprocess dataset train model. compared with state-of-the-art method based skeletal tracker cameras. Results show our achieves better performances avoids various drawbacks caused using tracker.Finally, also studied performance according different architectures dimensions minimize time required for acquisition