1/8x

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

Arthur Jakobsson1, Abhinav Mahajan1, Karthik Pullalarevu1, Krishna Suresh1, Yunchao Yao2, Yuemin Mao1, Bardienus Duisterhof1, Shahram Najam Syed1, Jeffrey Ichnowski1
1Momentum Lab, CMU Robotics Institute (Carnegie Mellon University), 2The University of North Carolina at Chapel Hill
Accepted to CoRL 2026
arXiv Video Thread

Wiggle and Go! uses a brief wiggle motion to identify the dynamic behavior of a rope, then goes—performing goal-conditioned manipulation in one shot.

3.55 cmmean 3D target striking error in real-world tests with learned rope parameters
0.95mean Pearson frequency correlation between simulated and real rope motion on held-out actions
1x
1/8x

Abstract

Many robotic tasks are unforgiving; a single mistake in a dynamic throw can lead to unacceptable delays or unrecoverable failure. We introduce Wiggle and Go!, a two-stage framework for zero-shot rope manipulation: a brief, safe wiggle action is observed to predict descriptive rope parameters, which then conditions a trajectory optimizer for zero-shot goal-conditioned execution. Unlike prior dynamic rope manipulation methods that require large real-world datasets or iterative real-world refinement, our identification module is task-agnostic, supporting diverse manipulation policies without retraining. We achieve a 3.55 cm average accuracy on 3D target striking in real using rope system parameters in comparison to 15.29 cm for uninformed baselines, and over 50% success on multi-objective lobbing and draping tasks. Predicted parameters transfer to unseen motions with 0.95 Pearson correlation between simulated and real rope dynamics, indicating that the identification module generalizes across the task corpus.

Research video

A brief wiggle identifies rope dynamics before zero-shot striking, lobbing, and draping. Watch on YouTube.

Pipeline Overview

Pipeline: a real robot wiggles a rope, a simulation-trained network estimates rope parameters, and a policy optimizer plans the goal-directed motion.
The Wiggle and Go! pipeline: We train a system identification neural network Φ-NN entirely in simulation. We perform a wiggle in real as an input for Φ-NN. The output of these parameters are used to optimize an action for a chosen task which is then executed in real. Items highlighted in green are the only elements of our procedure performed in real, once each.

Three dynamic rope tasks

Blender render of the xArm robot performing 3D rope striking, draping over a wall, and lobbing onto a target
Drake tasks rendered in Blender: 3D striking, draping, and lobbing. Green marks the extender; red and blue mark targets.

System Identification Intuition

This section shows the stacked videos of the varying rope parameters performed with the same wiggle motion. It helps us develop intuition about the rope behavior and how it changes with different parameters.

Lead Mass, Length Varied Ropes Stacked

Rope Varied; Length, Lead Mass Constant

Interactive Drake task animation

Explore the original Drake recording: drag to orbit, scroll to zoom, or use the playback controls to inspect the motion.

Full pipeline task completions

1/8x

Raw experiment videos Original camera recordings, organized by task and rope configuration.

Acknowledgments

This work used computational resources at Pittsburgh Supercomputing Center through allocation CIS251423 from the Advanced Cyberinfrastructure Coordination Ecosystem: Services & Support (ACCESS) program, which is supported by U.S. National Science Foundation grants #2138259, #2138286, #2138307, #2137603, and #2138296.

We also thank Professor Chris Atkeson for his advice on the paper and Alan Wang for his early work on this project.

Author's Note

I am a masters student in Robotics at Carnegie Mellon University. My fields of interest include: robot learning, manipulation and active perception. I am looking to pursue a PhD in Robotics and Machine Learning. If you have collaboration opportunities or any question about my work, please contact me at ajakobss+wag@cmu.edu.

BibTeX

@inproceedings{jakobsson2026wag,
  author    = {Jakobsson, Arthur and Mahajan, Abhinav and Pullalarevu, Karthik and Suresh, Krishna and Yao, Yunchao and Mao, Yuemin and Duisterhof, Bardienus and Syed, Shahram Najam and Ichnowski, Jeffrey},
  title     = {Wiggle and Go! System Identification for Zero-Shot Dynamic Rope Manipulation},
  booktitle = {10th Conference on Robot Learning},
  year      = {2026},
}