CableRobotGraphSim: A Graph Neural Network for Modeling Partially Observable Cable-Driven Robot Dynamics

Nelson Chen1, William R. Johnson III2, Rebecca Kramer-Bottiglio2, Kostas Bekris1, Mridul Aanjaneya1
1 Rutgers University     2 Yale University
arXiv · February 2026
Paper PDF Code BibTeX
CableRobotGraphSim method pipeline

Overview of the CableRobotGraphSim pipeline: graph construction from partial observations, GNN-based dynamics prediction, sim-and-real co-training, and closed-loop MPPI control.

Abstract

General-purpose simulators have accelerated the development of robots. Traditional simulators based on first-principles, however, typically require full-state observability or depend on parameter search for system identification. This work presents CableRobotGraphSim, a novel Graph Neural Network (GNN) model for cable-driven robots that aims to address shortcomings of prior simulation solutions. By representing cable-driven robots as graphs, with the rigid-bodies as nodes and the cables and contacts as edges, this model can quickly and accurately match the properties of other simulation models and real robots, while ingesting only partially observable inputs. Accompanying the GNN model is a sim-and-real co-training procedure that promotes generalization and robustness to noisy real data. This model is further integrated with a Model Predictive Path Integral (MPPI) controller for closed-loop navigation, which showcases the model's speed and accuracy.

Video

▶

Video coming soon

Replace this block with a <video> tag or YouTube embed

Overview of CableRobotGraphSim: graph construction, GNN forward passes, and closed-loop MPPI navigation. (Video has sound.)

Key Contributions

🕸️

Graph Representation

Rigid bodies as nodes, cables and contacts as edges. The graph topology mirrors physical structure, providing a strong inductive bias for learning dynamics.

👁️

Partial Observability

Operates on incomplete sensor data—no full state estimation or privileged simulation access is required during inference.

🔄

Sim-and-Real Co-Training

A joint training procedure leveraging abundant clean simulation data alongside scarce, noisy real robot rollouts to improve real-world generalization.

🎯

MPPI Integration

The learned forward model integrates directly with a Model Predictive Path Integral controller for fast, accurate closed-loop navigation on real hardware.

Input (t)
observed state
single inference
→
GNN
Output (t+h)
predicted state
Multi-Step Prediction. In a single forward pass, CableRobotGraphSim predicts the robot state h steps into the future directly from a partially-observed input at time t, no auto-regressive rollout required.

BibTeX

@article{chen2026cablerobotgraphsim,
  title   = {CableRobotGraphSim: A Graph Neural Network for
             Modeling Partially Observable Cable-Driven
             Robot Dynamics},
  author  = {Chen, Nelson and Johnson III, William R. and
             Kramer-Bottiglio, Rebecca and Bekris, Kostas and
             Aanjaneya, Mridul},
  journal = {arXiv preprint arXiv:2602.21331},
  year    = {2026}
}