Optimizing Multi-Touch Textile and Tactile Skin Sensing Through Circuit Parameter Estimation

Bo Ying Su, Yuchen Wu, Chengtao Wen, Changliu Liu

Abstract

Textile-based tactile skins offer a cost-effective and flexible solution for robotic sensing, yet they are frequently plagued by "ghosting"—the detection of phantom touches—and inaccuracies caused by the internal resistance of conductive threads. This paper presents a signal-processing framework that models textile skins as dynamic electrical circuits. By employing Regularized Least Squares to estimate individual cell resistances, the proposed method improves force sensing accuracy by approximately 27% and effectively mitigates ghosting. While computational scalability remains a challenge for high-resolution arrays, this approach demonstrates that sophisticated software modeling can compensate for simplified, high-sensitivity hardware designs.

Introduction

Tactile sensing is a critical component in enabling robots to interact safely and effectively with their environments. Among various sensing modalities, textile-based skins are particularly promising due to their low cost, breathability, and ability to conform to the complex geometries of robotic limbs. However, these skins typically suffer from a phenomenon known as "ghosting." In a standard resistive grid, current often follows unintended paths—the path of least resistance—leading the system to report touches in locations where no physical contact has occurred. Furthermore, the conductive threads used in these textiles possess significant internal resistance that fluctuates as the fabric stretches, complicating traditional voltage-to-force calculations.

Methodology

To address these challenges, this research shifts the complexity from physical hardware to mathematical optimization. The approach consists of a two-layer hardware design paired with a robust circuit estimation pipeline.

Hardware Design

Rather than utilizing complex three-layer architectures designed to physically isolate touch points, the researchers adopted a simplified two-layer design featuring orthogonally knitted conductive stripes. While this configuration is inherently more prone to ghosting, it offers superior sensitivity to light touches (detecting forces as low as 1N) and is significantly easier to manufacture.

Circuit Parameter Estimation

The core of the solution lies in treating the entire skin as a resistive sensory array. The system solves for the unknown resistances ($R^C$) of every cell in the grid using a Regularized Least Squares objective function. This optimization is handled by the Ipopt solver, an interior-point algorithm designed to find the best-fitting circuit parameters that explain observed voltage drops.

To ensure the model remains physically grounded, an $L_2$ regularization term is included. This penalizes implausible wire resistances, forcing the solver to attribute voltage changes to actual physical touches rather than noise or minor fabric deformations.

Operational Pipeline

The framework operates in two distinct stages:

  1. Calibration: A robotic arm performs a controlled single-point touch to establish a linear regression model for each individual cell.
  2. Estimation: During real-time operation, the system solves the optimization problem to predict forces across the grid, even during simultaneous multi-touch events.

Experimental Evaluation

The framework was validated through both simulation and physical experimentation. Circuit simulations were conducted using the Lcapy tool to test the optimization algorithm against various resistance scenarios. Real-world data was collected using a Kinova Gen3 robot equipped with a Robotiq FT 300-S Force Torque Sensor. Testing was performed on both flat and curved surfaces to simulate the varied surfaces of a robot's body.

Results and Discussion

The experimental results demonstrate a significant performance leap over "naive" estimation methods:

  • Accuracy Improvements: The proposed method reduced the Root Mean Square Error (RMSE) in force prediction by 27.3% on flat surfaces and 26.4% on curved surfaces.
  • Ghosting Mitigation: Time-series analysis confirmed that the algorithm successfully isolated individual touches. By accounting for the circuit's global behavior, the system significantly reduced the "correlated noise" in adjacent cells that typically characterizes ghosting.
  • Real-Time Viability: On an AMD 7940HS CPU, the system achieved a readout rate of 11 samples per second for a 2x2 grid, confirming its utility for real-time interaction in small-scale applications.

Conclusion and Future Work

This research proves that the inherent limitations of textile-based tactile sensors—specifically ghosting and internal resistance—can be largely overcome through sophisticated software modeling. By treating the skin as a circuit rather than a collection of independent sensors, the system achieves higher accuracy and sensitivity without the need for expensive or rigid hardware components.

The primary limitation of the current framework is scalability; the computational cost of non-linear programming increases significantly with the size of the sensor grid. Future research will focus on optimizing solver efficiency and exploring machine learning techniques to approximate optimization results, enabling high-resolution tactile skins to operate at real-time speeds.

Summary updated

BibTeX
@inproceedings{su2024optimizing,
  author = {Bo Ying Su and Yuchen Wu and Chengtao Wen and Changliu Liu},
  title = {{Optimizing Multi-Touch Textile and Tactile Skin Sensing Through Circuit Parameter Estimation}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2024},
  doi = {10.1109/icra57147.2024.10610053},
}
Figure