RopeDreamer: A Kinematic Recurrent State Space Model for Dynamics of Flexible Deformable Linear Objects
2026-04-30 • Robotics
Robotics
AI summaryⓘ
The authors address the challenge of predicting how flexible objects like ropes or cables (Deformable Linear Objects) move and change shape when manipulated by robots. They developed a new method that represents these objects using rotations (quaternions) instead of simple positions, helping to keep the model physically accurate and prevent unrealistic stretching or tangling. Their approach uses a specialized neural network design to better capture the object's behavior over time and tested it on complex simulated tasks. The results show their model predicts future states more accurately and faster than existing methods, while also keeping the object's shape consistent during complicated movements.
Deformable Linear ObjectsRobotic ManipulationRecurrent State Space ModelQuaternion RepresentationLatent DynamicsGraph Neural NetworksTopological IntegritySelf-intersectionsOpen-loop PredictionKinematic Chain
Authors
Tim Missal, Lucas Domingues, Berk Guler, Simon Manschitz, Jan Peters, Paula Dornhofer Paro Costa
Abstract
The robotic manipulation of Deformable Linear Objects (DLOs) is a fundamental challenge due to the high-dimensional, non-linear dynamics of flexible structures and the complexity of maintaining topological integrity during contact-rich tasks. While recent data-driven methods have utilized Recurrent and Graph Neural Networks for dynamics modeling, they often struggle with self-intersections and non-physical deformations, such as tangling and link stretching. In this paper, we propose a latent dynamics framework that combines a Recurrent State Space Model with a Quaternionic Kinematic Chain representation to enable robust, long-term forecasting of DLO states. By encoding the DLO as a sequence of relative rotations (quaternions) rather than independent Cartesian positions, we inherently constrain the model to a physically valid manifold that preserves link-length constancy. Furthermore, we introduce a dual-decoder architecture that decouples state reconstruction from future-state prediction, forcing the latent space to capture the underlying physics of deformation. We evaluate our approach on a large-scale simulated dataset of complex pick-and-place trajectories involving self-intersections. Our results demonstrate that the proposed model achieves a 40.52% reduction in open-loop prediction error over 50-step horizons compared to the state-of-the-art baseline, while reducing inference time by 31.17%. Our model further maintains superior topological consistency in scenarios with multiple crossings, proving its efficacy as a compositional primitive for long-horizon manipulation planning.