Wiggle and Go! System Identification for Zero-Shot Dynamic Rope Manipulation

Carnegie Mellon University researchers have developed Wiggle and Go!, a novel system-identification framework that enables robots to dynamically manipulate ropes with zero-shot learning—no real-world training data required. The method addresses a critical challenge in robotics: tasks like throwing a rope to a target are unforgiving, where a single mistake causes delays or failure. Traditional approaches either need large real-world datasets or iterative trial-and-error. Wiggle and Go! instead uses a two-stage pipeline: first, a system identification module observes rope movement to predict descriptive physical parameters (like stiffness and damping). Second, these parameters inform a goal-conditioned optimization method that predicts robot actions for zero-shot execution in the real world.

In experiments, the method achieved 3.55 cm average accuracy on 3D target striking when using system parameters, versus 15.34 cm without them—a 4x improvement. It also reached a 0.95 Pearson correlation coefficient between Fourier frequencies of predicted and real ropes on unseen trajectories, showing high fidelity in modeling rope dynamics. The task-agnostic system ID module allows seamless switching between different manipulation tasks, supporting diverse policies from a single model. This work, submitted to arXiv on April 23, 2026, represents a significant step toward practical robotic manipulation of deformable objects in dynamic environments.