BoxTwin: Learning Elastoplastic Articulated Object Dynamics from Videos
Robotics
Summary
The authors developed BoxTwin, a digital twin system that helps robots understand and predict how bendy, foldable objects move and change shape when handled. By learning from videos, their system builds a model of these objects that includes how they stretch, bend, and even get permanently deformed. They showed that BoxTwin can accurately follow joint movements and predict lasting changes after the objects are touched or folded. This work moves digital twins closer to controlling soft, jointed objects in messy, real-world settings.
Authors
Heng Zhang, Gehan Zheng, Kaifeng Zhang, Jay Song, Shivansh Patel, Sonny Hu, Yunzhu Li, Changxi Zheng, Peter Yichen Chen
Abstract
Digital twins enable robots to anticipate and adapt to physical interactions, but existing models struggle with elastoplastic articulated objects (EAOs) that exhibit nonlinear elasticity, plastic yielding, and damage accumulation. We present BoxTwin, an interactive digital twin framework that learns the full dynamics of EAOs from videos. Our pipeline reconstructs the scene, identifies a physics aware constitutive model for each EAO. Experiments on manual folding and dual arm manipulation of EAOs show that BoxTwin accurately tracks joint trajectories and reproduces post contact plastic behavior over long horizons. By integrating video driven reconstruction with elastoplastic damage modeling, BoxTwin advances digital twins toward predictive, adaptive control of deformable articulated objects in unstructured environments.