Neural Network-Based Active Load-Sensing Scheme and Stiffness Adjustment for Pneumatic Soft Actuators for Minimally Invasive Surgery Support

Yuxi Lu, Zhongchao Zhou, Shota Kokubu, Ruian Qin, Pablo E. Tortós Vinocour, Wenwei YuSensors 2023 · DOI

TL;DR

This paper introduces a novel neural network (NN)-based system for soft pneumatic actuators that can actively sense external loads and adjust their stiffness. This is crucial for supporting flexible endoscopes in minimally invasive surgery (MIS) by providing a stable view and preventing damage to tissues. The system uses an "environment exploration strategy" to improve load estimation accuracy and an inverse dynamics model to control stiffness.

What Problem Does This Paper Address?

Minimally invasive surgery (MIS) relies on flexible endoscopes to navigate complex internal anatomy. However, these endoscopes can be easily deformed by external forces from surrounding organs, leading to an unstable surgical view and potential tissue damage. While soft actuators offer a promising solution for providing adjustable stiffness, accurately estimating the external forces acting on them is challenging due to their inherent nonlinearities and the complex interactions with their environment. Existing methods for load sensing are often incompatible with soft actuators or lack the robustness needed for real-world surgical scenarios. This paper aims to bridge this gap by developing a robust and accurate method for load sensing and stiffness control in soft actuators for MIS applications.

Methods

The core of the proposed solution involves two main components, both leveraging neural networks:

  • Active Load-Sensing Scheme with Environment Exploration:
    • A neural network (NN) is trained to estimate the external load based on input features like air pressures in antagonistic chambers (P_FU, P_FD) and the initial and final bending angles of the soft actuator.
  • To enhance robustness against the nonlinearities of the soft actuator and environmental uncertainties, an "environment exploration strategy" is introduced. This involves slightly perturbing the air pressure in one of the chambers (P_FD) and observing the resulting changes. The average of the estimated loads over these perturbations is used as the final load estimate.
  • Different "exploration sequences" (varying lengths and directions of pressure changes) were tested to find the optimal strategy for improving accuracy and reducing variance.
  • NN-Based Inverse Dynamics Model for Stiffness Adjustment:
    • Once the external load is estimated, a second NN is used to determine the required air pressure (P_FD) to achieve a desired stiffness.
  • This NN acts as an inverse dynamics model, taking the desired stiffness, estimated external load, endoscope diameter, and final bending angle as inputs.

Datasets

The study utilized experimental data collected from a custom-built soft pneumatic actuator. The data was generated by varying:

  • Air pressures in the antagonistic chambers (P_FU and P_FD).
  • The type of endoscope inserted (diameters of 3.8 mm, 5.0 mm, and 5.4 mm).
  • Applied external loads (ranging from 0 N to 1.06 N).

This data was split into training and verification sets. The training data included loads of 0, 0.17, 0.27, 0.56, and 1.06 N for all endoscope diameters. Verification data used loads of 0.15 and 0.22 N for the 5 mm endoscope.

Key Insights and Ideas

  • Exploration is Key for Robustness: The "environment exploration strategy" is a critical innovation. Simply relying on a single measurement is insufficient due to the complex and nonlinear nature of soft actuators. By actively probing the environment, the system can average out noise and uncertainties, leading to more reliable load estimations.
  • The Power of Averaging: The exploration strategy effectively acts as a form of moving average, smoothing out transient errors and providing a more stable estimate of the external load.
  • Endoscope Diameter Matters: The diameter of the inserted endoscope significantly influences the soft actuator's bending angle and overall stiffness, highlighting the need to account for this variable in any control strategy.
  • Nonlinearity is the Norm: The relationship between air pressure, bending angle, and stiffness in soft actuators is highly nonlinear, making traditional analytical modeling difficult. Neural networks are well-suited to capture these complex relationships.

Key Results

  • Improved Load Estimation Accuracy: The environment exploration strategy significantly improved the accuracy of external load estimation compared to a single-shot instantaneous load estimation. The successful estimation rate reached up to 80% for certain exploration sequences.
  • Optimal Exploration Sequence Identified: The exploration sequence labeled "synthesized load (±2)" was found to be optimal, offering a good balance between accuracy, reduced variance, and efficient exploration time.
  • Effective Stiffness Control: The NN-based inverse dynamics model demonstrated its ability to accurately calculate the required air pressure (P_FD) to achieve a desired stiffness, even with the nonlinear relationship between pressure and stiffness.
  • Performance Below 35 kPa: The stiffness adjustment model showed limitations when the true P_FD was below 35 kPa, indicating that at very low pressures, the stiffness variation is minimal, making accurate prediction challenging.

Limitations and Future Work

  • Limited Pressure Range for Stiffness Control: The inverse dynamics model struggled to accurately predict air pressure for desired stiffness when the true P_FD was below 35 kPa.
  • Generalizability of Optimal Sequence: The identified optimal exploration sequence (±2) was specific to the experimental setup and soft actuator used. Further research is needed to assess its generalizability across different soft actuators and configurations.
  • Multi-DoF Expansion: The current work focused on a single degree of freedom (DoF) for stiffness control. Future work should expand the active load-sensing scheme to multi-DoF soft actuators.
  • Closed-Loop Control: The paper suggests designing a closed-loop feedback controller for desired angle and stiffness, which would represent a significant advancement in autonomous control for MIS.

Summary updated

Abstract

To provide a stable surgical view in Minimally Invasive Surgery (MIS), it is necessary for a flexible endoscope applied in MIS to have adjustable stiffness to resist different external loads from surrounding organs and tissues. Pneumatic soft actuators are expected to fulfill this role, since they could feed the endoscope with an internal access channel and adjust their stiffness via an antagonistic mechanism. For that purpose, it is essential to estimate the external load. In this study, we proposed a neural network (NN)-based active load-sensing scheme and stiffness adjustment for a soft actuator for MIS support with antagonistic chambers for three degrees of freedom (DoFs) of control. To deal with the influence of the nonlinearity of the soft actuating system and uncertainty of the interaction between the soft actuator and its environment, an environment exploration strategy was studied for improving the robustness of sensing. Moreover, a NN-based inverse dynamics model for controlling the stiffness of the soft actuator with different flexible endoscopes was proposed too. The results showed that the exploration strategy with different sequence lengths improved the estimation accuracy of external loads in different conditions. The proposed method for external load exploration and inverse dynamics model could be used for in-depth studies of stiffness control of soft actuators for MIS support.
BibTeX
@inproceedings{lu2023neural,
  author = {Yuxi Lu and Zhongchao Zhou and Shota Kokubu and Ruian Qin and Pablo E. Tortós Vinocour and Wenwei Yu},
  title = {{Neural Network-Based Active Load-Sensing Scheme and Stiffness Adjustment for Pneumatic Soft Actuators for Minimally Invasive Surgery Support}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s23020833},
}
Figure