Learning Friction Model for Magnet-Actuated Tethered Capsule Robot

Yi Wang, Yuyang Tu, Yuchen He, Xutian Deng, Ziwei Lei, Jianwei Zhang, Miao Li

TL;DR

This paper introduces a novel learning-based approach to model friction for magnet-actuated tethered capsule robots, a crucial step for improving their diagnostic and therapeutic capabilities. By developing a dynamic friction model and experimentally validating it across various contact surfaces, the authors demonstrate significant improvements in trajectory tracking accuracy and contact force stability.

What Problem Does This Paper Address?

Magnet-actuated tethered capsule robots hold great promise for minimally invasive medical procedures like gastrointestinal inspections. However, achieving precise control of these robots is challenging due to unpredictable forces like friction between the capsule and the internal environment, as well as drag from the tether. Traditional methods struggle to accurately model these forces, especially when the contact medium changes. This paper tackles the problem of accurately modeling and compensating for friction in these systems to enhance their control accuracy and reliability.

Methods

The authors propose a learning-based approach to model the friction experienced by a tethered capsule robot. The core of their method involves:

  • Dynamic Model Formulation: They start with a dynamic model of the magnetron capsule, incorporating magnetic forces and torques.
  • Friction Model Development: A rate-dependent friction model is proposed, where the friction coefficient is a function of the capsule's velocity and a state variable. This model is then simplified for practical application.
  • Experimental Setup: A system comprising an external permanent magnet (EPM) mounted on a robotic arm and an internal permanent magnet (IPM) within the tethered capsule robot is designed. Experiments are conducted on a transparent PVC plate, paper, and polyester-cloth to simulate different contact environments.
  • Data Acquisition: The velocities of the EPM and IPM, along with the capsule's acceleration, are measured. Magnetic forces (F and Fz) are calculated using finite element analysis.
  • Friction Parameter Learning: The relationship between velocity and the friction factor (c) is learned by fitting experimental data using a fifth-degree polynomial. This allows for the determination of specific friction models for different contact mediums.
  • Control Integration: The learned friction model is integrated into the control system to compensate for friction during motion.

Datasets

The paper does not explicitly mention pre-existing datasets. Instead, it describes the creation of a new dataset through experimental procedures. This dataset consists of:

  • Trajectory Tracking Data: Coordinates of the manipulator's end (EPM) and the capsule robot (IPM) collected at 20Hz.
  • Force and Acceleration Data: Measured drag force, calculated magnetic forces (F and Fz), and capsule acceleration.
  • Contact Mediums: Experiments were conducted on PVC, paper, and polyester-cloth surfaces, providing data for three distinct friction conditions.

Key Insights and Ideas

  • Friction is a Dominant Factor: The paper strongly emphasizes that friction, often overlooked or simplified, plays a critical role in the motion control of tethered capsule robots and significantly impacts their accuracy.
  • Learning for Adaptability: A learning-based approach is crucial for adapting to varying friction conditions encountered in different biological environments.
  • Velocity-Dependent Friction: The friction model's dependence on velocity is a key insight, moving beyond static friction models to capture more realistic behavior.
  • Universality of the Method: The experimental validation across three different contact mediums highlights the potential universality of their friction modeling approach.
  • Practical Implications: By accurately modeling friction, the authors show a clear path to improving trajectory tracking and contact force stability, which are essential for safe and effective medical procedures.

Key Results

  • Improved Trajectory Tracking: The learned friction model led to a significant reduction in trajectory tracking error. For example, on PVC, the average accuracy improvement for different speeds (V1, V2, V3) ranged from 4.16% to 6.01% compared to a baseline.
  • Enhanced Contact Force Stability: The paper reports an improvement in contact force stability, with a 6.12% and 6.47% improvement on PVC when using the learned friction model compared to not using it.
  • Accurate Friction Modeling: The fifth-degree polynomial fitting for the friction factor 'c' achieved a small error (0.053) on PVC within the tested speed range.
  • Validation Across Mediums: The method was successfully applied to PVC, paper, and polyester-cloth, demonstrating its applicability to environments with varying friction coefficients.

Limitations and Future Work

  • Experimental Environment: The current experiments are conducted on a transparent PVC plate, paper, and polyester-cloth, which are simplified representations of the complex biological environment. The authors acknowledge that the "contact force in current experimental scene is nearly a simulation to the actual working process of the capsule."
  • Compact Structure: The current system's structure is not optimized for size. Future work aims to design a more compact system.
  • Force Estimation: The authors plan to integrate force estimation into the system to facilitate more advanced force control.
  • Animal Tissue Testing: Further verification of the friction model in actual animal tissues is suggested as a crucial next step.

Summary updated

Abstract

The potential diagnostic applications of magnet-actuated capsules have been greatly increased in recent years. For most of these potential applications, accurate position control of the capsule have been highly demanding. However, the friction between the robot and the environment as well as the drag force from the tether play a significant role during the motion control of the capsule. Moreover, these forces especially the friction force are typically hard to model beforehand. In this paper, we first designed a magnet-actuated tethered capsule robot, where the driving magnet is mounted on the end of a robotic arm. Then, we proposed a learning-based approach to model the friction force between the capsule and the environment, with the goal of increasing the control accuracy of the whole system. Finally, several real robot experiments are demonstrated to showcase the effectiveness of our proposed approach.
BibTeX
@inproceedings{wang2022learning,
  author = {Yi Wang and Yuyang Tu and Yuchen He and Xutian Deng and Ziwei Lei and Jianwei Zhang and Miao Li},
  title = {{Learning Friction Model for Magnet-Actuated Tethered Capsule Robot}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2022},
  doi = {10.1109/icra46639.2022.9811587},
}
Figure