Capability Verification and Error Source Investigation of External Force Sensing Methods of Continuum Robot

AnFan Zhang, Bowen Yan, Xueqian Wang, Bin Liang, Zhiheng LiResearch Square 2022 · DOI

Abstract

This paper studies mainly error sources of current external force sensing algorithms and presents external force sensing algorithms of continuum robots using superelastic NiTi backbones. Axial compression is deliberately taken into consideration in analytical dynamics for facilitating the later analysis of error sources. The analytical dynamics of multibackbone continuum robots is established by Lagrange equation in a simple and concise manner. Experiments show that the force-strain relationship is significantly different with or without external force, and we provide some insights into the feasibility and limitations of dynamic model. Then an extend intrinsic force sensing algorithm (EIFS) and an external force observer based on generalized momentum method (EFOGM) is developed, Both of them only depend on joint-level information. The effectiveness of both force sensing algorithms is validated through simulations and experiments. Mean absolute error of EIFS is 0.1903 N using constant axial strain and 0.0866 N using real-time axial strain, respectively. While mean absolute errors of EFOGM is 0.1908 N using constant axial strain and 0.0293 N using real-time axial strain, respectively. The estimation effect of EFOGM is slightly preferable to that of EIFS. Experiments shows that the main sources of force sensing algorithm error are: 1) the estimated external force compensates the force part caused by axial compression, 2) model error from nonlinear stress-strain relationship of cable tension-induced deformation and external force-induced deformation under the loading and unloading force stage.

Introduction

Continuum robots—flexible, snake-like manipulators—are increasingly utilized in minimally invasive surgery due to their ability to navigate complex anatomical structures. However, safe interaction with delicate tissues requires precise knowledge of external forces. Traditional force sensors are often unsuitable for these applications because they are bulky, difficult to sterilize, and sensitive to electromagnetic interference.

"Sensorless" force sensing offers an alternative by estimating external forces through mathematical models of the robot's internal state. However, existing models frequently suffer from "ghost" errors because they overlook critical physical phenomena such as axial compression and non-linear material stiffness. This research addresses these gaps by refining the analytical dynamics of multi-backbone continuum robots to improve force estimation accuracy.

Modeling and Methodology

The researchers established a comprehensive dynamic model using Lagrange equations, explicitly accounting for axial compression of the primary backbone, friction between driving cables and internal discs, and non-linear material properties. Specifically, they addressed the superelastic nature of the Nickel-Titanium (NiTi) backbone by introducing a corrected Young’s Modulus ($E$) that varies with the deflection angle.

Geometric model of the multi-backbone continuum robot.

The study compared two primary algorithms:

  1. Extended Intrinsic Force Sensing (EIFS): Based on the principle of virtual work, mapping joint-level information to external forces.
  2. External Force Observer based on Generalized Momentum (EFOGM): A dynamic observer that estimates forces without requiring noisy acceleration measurements.

Experimental validation was conducted using one-segment continuum robots with varying disc counts (3 and 7). Tip positions were tracked using an OptiTrack motion capture system, while actual forces were measured using tension gauges and calibration weights ranging from 100g to 1500g.

Results and Discussion

The investigation revealed that axial strain is a primary source of error in force sensing. If a model assumes constant backbone length, any compression under load is misinterpreted as an external force. Incorporating real-time axial strain into the algorithms led to a significant reduction in estimation error.

Furthermore, the study demonstrated that the Young’s Modulus of NiTi backbones is not constant. As the deflection angle ($\beta$) increases, the effective stiffness changes. Implementing a piecewise correction for $E$ significantly improved the model's fidelity at high deflection angles.

Comparison of experimental and theoretical force-deflection relationships for 3-disc and 7-disc robots.

The EFOGM algorithm consistently outperformed EIFS, particularly when real-time axial strain was utilized. The mean absolute error for EFOGM dropped from 0.1908 N (with constant axial strain) to just 0.0293 N (with real-time axial strain). Additionally, increasing the number of discs from 3 to 7 helped maintain a more predictable robot shape, further enhancing model accuracy.

Conclusion

This work highlights the critical importance of accounting for axial compression and non-linear material properties in the force sensing of continuum robots. By refining the dynamic models and employing the EFOGM observer, the researchers achieved high-precision force estimation using only internal state data. Future research will focus on extending these methods to multi-segment robots and addressing material memory effects, such as residual strain during loading and unloading cycles.

Summary updated

Abstract

<title>Abstract</title> This paper studies mainly error sources of current external force sensing algorithms and presents external force sensing algorithms of continuum robots using superelastic NiTi backbones. Axial compression is deliberately taken into consideration in analytical dynamics for facilitating the later analysis of error sources. The analytical dynamics of multibackbone continuum robots is established by Lagrange equation in a simple and concise manner. Experiments show that the force-strain relationship is significantly different with or without external force, and we provide some insights into the feasibility and limitations of dynamic model. Then an extend intrinsic force sensing algorithm (EIFS) and an external force observer based on generalized momentum method (EFOGM) is developed, Both of them only depend on joint-level information. The effectiveness of both force sensing algorithms is validated through simulations and experiments. Mean absolute error of EIFS is 0.1903 N using constant axial strain and 0.0866 N using real-time axial strain, respectively. While mean absolute errors of EFOGM is 0.1908 N using constant axial strain and 0.0293 N using real-time axial strain, respectively. The estimation effect of EFOGM is slightly preferable to that of EIFS. Experiments shows that the main sources of force sensing algorithm error are: 1) the estimated external force compensates the force part caused by axial compression, 2) model error from nonlinear stress-strain relationship of cable tension-induced deformation and external force-induced deformation under the loading and unloading force stage.
BibTeX
@inproceedings{zhang2022capability,
  author = {AnFan Zhang and Bowen Yan and Xueqian Wang and Bin Liang and Zhiheng Li},
  title = {{Capability Verification and Error Source Investigation of External Force Sensing Methods of Continuum Robot}},
  booktitle = {Research Square},
  year = {2022},
  doi = {10.21203/rs.3.rs-1413484/v1},
}

Loading related content...

Figure