Adaptive Force-field Control of a 2-DOF Upper-extremity Rehabilitation Robot

Parya Khoshroo, Behzad Danaei, John McPhee, Jennifer BogerProceedings of the International Conference of Control, Dynamic Systems, and Robotics 2020 · DOI

Introduction

Stroke survivors frequently face a long and arduous journey toward regaining motor function, often requiring intensive, repetitive physical therapy. While robotic systems have been introduced to automate these tasks and increase the frequency of therapy, a significant challenge remains: personalization. Every patient presents a unique profile of motor deficits; some exhibit poor muscle coordination leading to erratic movements, while others suffer from a lack of strength that prevents them from reaching targets.

Researchers at the University of Waterloo are addressing this by developing an adaptive control algorithm for rehabilitation robots. Their work explores the balance between assistive and resistive forces, investigating how robotic intervention can best facilitate neural adaptation and long-term motor recovery.

The Personalization Problem in Robotic Therapy

Current robotic therapies often rely on a "one-size-fits-all" approach or focus primarily on assisting the patient to complete a movement. However, a phenomenon known as the "slacking" effect suggests that if a robot performs too much of the work, the patient’s brain may not undergo the necessary neural adaptation required to relearn the movement. To combat this, the researchers propose an adaptive system that can independently adjust the intensity of assistance or resistance based on a patient's evolving performance.

Methodology

The 2-DOF Manipulandum and Force-Field Strategy

The study focuses on a 2-Degree of Freedom (DOF) manipulandum—a robotic arm that a patient moves across a horizontal plane. The core of the control strategy involves two distinct types of force fields:

  1. Lateral Force Fields: These apply forces perpendicular to the desired movement path. They are specifically designed to target coordination issues (Category 1 impairments), pushing or pulling the hand to correct shaky or off-course movements.
  2. Radial Force Fields: These apply forces toward or away from the goal. These target strength-related deficits (Category 2 impairments), helping or hindering the user's ability to reach the target.

The algorithm is designed to be adaptive, meaning it can concurrently but independently tune the intensity of these two fields in real-time. This allows for a highly customized therapy session where a patient might be challenged in their coordination while being assisted in their strength.

Dynamics and Simulation Modeling

To derive the robot's dynamics, the team utilized the Principle of Virtual Work. This approach is computationally more efficient than standard Newton-Euler methods for this application because it eliminates internal forces and constraint equations from the outset, simplifying the real-time control calculations.

To validate the algorithm without clinical risk, the researchers employed a sophisticated human motor adaptation model based on the scheme proposed by Shadmehr and Mussa-Ivaldi (1994). They modified this model to simulate post-stroke impairments by adding a "biased step disturbance" and capping the maximum hand force. This created a "virtual stroke patient" with controllable levels of impairment.

Key Insights: The "No Pain, No Gain" Reality

The simulation results revealed a striking contrast between performance during training and actual motor recovery:

  • The Assistive Mode: When the robot provided an assistive force field, the virtual patient performed the task nearly perfectly. However, this success was illusory; once the assistance was removed, performance dropped significantly. The brain essentially "slacked off" because the robot was compensating for the impairment.
  • The Resistive Mode: Conversely, resistive force fields—which actively pushed the patient away from the goal or off the path—resulted in the worst performance during the training session. However, this struggle acted as a catalyst for adaptation. Forced to overcome the resistance, the simulated motor system showed the most significant improvement in natural movement once the force fields were removed.

These findings support the error-augmentation theory in motor learning, which suggests that making errors more apparent, rather than hiding them with assistance, is the key to triggering the brain's plastic response.

Conclusion and Future Work

This study demonstrates that while assistive robots can make a patient feel successful in the moment, resistive challenges are likely more effective for long-term rehabilitation. The ability to independently tune lateral and radial forces provides a novel framework for personalized therapy.

The primary limitation of this work is its reliance on simulation. While the human motor model is well-regarded, biological systems are inherently unpredictable. Future work will involve implementing this adaptive algorithm on a physical 2-DOF robot for clinical trials with post-stroke subjects. Additionally, the researchers aim to refine the algorithm to switch between assistive and resistive modes based on real-time indicators of patient fatigue and frustration.

Summary updated

BibTeX
@inproceedings{khoshroo2020adaptive,
  author = {Parya Khoshroo and Behzad Danaei and John McPhee and Jennifer Boger},
  title = {{Adaptive Force-field Control of a 2-DOF Upper-extremity Rehabilitation Robot}},
  booktitle = {Proceedings of the International Conference of Control, Dynamic Systems, and Robotics},
  year = {2020},
  doi = {10.11159/cdsr20.138},
}

Loading related content...

Figure