CableSense: MuJoCo Simulation-Guided Neural Networks for Force Estimation in Cable-Driven Manipulators
states are recorded. (c) Force perception network uses D cable tension inputs to estimate external force.
AI summary
Problem
Cable-driven manipulators lack integrated force sensors due to space and weight constraints, making safe environmental interaction difficult. Traditional model-based estimation methods are computationally heavy and sensitive to modeling inaccuracies, while existing data-driven approaches struggle with sim-to-real transfer.
Approach
The authors build a high-fidelity MuJoCo simulation model calibrated to a physical cable-driven arm to generate diverse force-tension datasets, then train a multi-task neural network that maps cable tension differences to external force location, magnitude, and direction.
Key results
- High-fidelity MuJoCo simulation model accurately replicates physical cable dynamics
- Two-stage neural network architecture effectively decouples contact point identification from force regression
- Over 98% accuracy in contact location estimation with 5.96° mean absolute direction error
- Successful sim-to-real transfer validated through consistent physical tension trends and preliminary hardware experiments
Why it matters
Enables safe, low-cost, and lightweight force perception for cable-driven robots without adding bulky sensors, advancing their deployment in constrained and interactive environments.
Abstract
Cable-driven serial manipulator (CDSM) has ad- vantages of lightweight structure, high flexibility, and inher- ent safety, making it suitable for operations in constrained spaces. However, interaction with the environment is inevitable. To address this limitation, we propose CableSense, a novel force-sensing approach that leverages actuation cable tension information exclusively, thereby eliminating the requirement for additional contact sensors. We first develop a high-fidelity MuJoCo simulation model based on the physical system, reduc- ing the sim-to-real gap through careful calibration of physical and mechanical parameters. Leveraging this simulation model, we generate a comprehensive dataset encompassing diverse external force scenarios. We then implement a multi-task deep learning framework CableSense, for both single-point and multi-point force identification. Experiments demonstrate that This work was supported in part by the National Natural Science Foundation of China under Grant 52405028, in part by the National Key R&D Program of China (2022YFB4701400/4701402), in part by the China Postdoctoral Science Foundation under Grant Number 2024M761635. (Corresponding authors: Yanan Li; Xueqian Wang). Chunru Yang, Xinruo Xu, Zhongrui Cui, Yanan Li and Xueqian Wang are with Shenzhen International Graduate School, Tsinghua University, Shenzhen 518055, China. (e-mail: ycr24@mails.tsinghua.edu.cn; xuxr24@mails.tsinghua.edu.cn; cuizr24@mails.tsinghua.edu.cn; yananli@sz.tsinghua.edu.cn; wang.xq@sz.tsinghua.edu.cn). CableSense achieves over 98% accuracy in contact location estimation, maintaining a mean absolute direction error of 5.96◦.