Answers are grounded in this paper and cite their sources.
Accurate yet computationally efficient models of soft-material deformation are essential for real-time prediction and control in robotic manipulation. Recent advances in machine learning for object detection and image segmentation have made it possible to reliably extract geometric observations from raw RGB data, opening new avenues for vision-based material modeling. Building on these advances, this work presents a vision-informed framework that couples coarse finite element simulations with continuous online parameter estimation to obtain action-responsive models of soft-material flow suitable for model-based planning and control. Leveraging RGB observations of squeeze flow experiments, the framework repeatedly identifies effective viscosity and slip parameters of a Stokes model over short moving windows via Gauss--Newton updates. Synthetic datasets and physical experiments demonstrate that this approach achieves reliable short-horizon predictions and maintains geometric consistency despite observation noise and partial parameter identifiability.
No takes yet. Share an insight, caveat, or question.
Weerathunge et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: