Robotic Rehabilitation Through Multilateral Shared Control Architecture

Srikar Annamraju, Harris Nisar, Anne Horowitz, Dusan M. StipanovicRobotics 2025 · DOI

TL;DR

This paper introduces a novel multilateral shared control architecture for robotic rehabilitation, enhancing patient recovery by providing full force feedback to both therapist and patient, and dynamically adjusting assistance based on patient progress. This system aims to overcome limitations in existing robotic rehabilitation platforms, leading to more effective and engaging therapy.

What Problem Does This Paper Address?

The rehabilitation of stroke patients, particularly those with upper extremity hemiparesis, faces significant challenges. A primary issue is the shortage of qualified therapists, leading to insufficient therapy dosage. Additionally, patients often experience a lack of motivation during traditional therapy, hindering their recovery. While robotic rehabilitation platforms offer a promising solution, current systems often suffer from complex architectures, limited force feedback, and slow learning rates. This paper addresses these limitations by proposing a simplified, yet more effective, multilateral shared control architecture that prioritizes full force feedback, enhanced learning, and dynamic patient performance monitoring.

Methods

The core of this paper is the development of a multilateral leader-follower architecture for robotic rehabilitation. Unlike traditional bilateral systems, this architecture involves two "leader" robots (one for the therapist, one for the patient) and a single "follower" robot representing the virtual environment.

Key methodological contributions include:

  • Simplified Multilateral Architecture: A five-channel architecture is proposed, reducing complexity compared to prior six-channel systems. This architecture ensures that both the therapist and patient receive full force feedback from the virtual environment.
  • Recovery-Based Authority: The system dynamically adjusts the patient's control authority over the virtual robot based on their "recovery level" (r), a measure of their progress. As the patient recovers, their control authority increases, while the therapist's assistance decreases.
  • Full Force Feedback: A significant departure from many existing systems, this architecture provides uncompromised force feedback from the virtual environment to both the therapist and the patient. This aims to create a more realistic and engaging experience.
  • Hybrid Controllers: Controllers are developed for the virtual robot (position controller) and the leader robots (hybrid position-force controllers) to ensure system stability and meet performance objectives, even with full force feedback. Passivity analysis is used to prove system stability.
  • Attribute-Wise Recovery: A novel concept is introduced where recovery is not just a single metric but can be broken down into specific attributes like amplitude, frequency, and tremor. This allows for more targeted training and faster error correction.

Datasets

The paper does not explicitly mention training datasets. For evaluation, the authors utilized a custom experimental setup involving Touch Devices (3D Systems®) robots. They also performed a comparison with a traditional architecture (specifically referencing [16]) on the same experimental testbed under similar conditions. The "dataset" in this context is the data collected from the experiments with human subjects operating the robotic system.

Key Insights and Ideas

  • Full Force Feedback is Crucial for Engagement and Learning: The paper strongly advocates for and demonstrates the benefits of providing complete force feedback to both users. This enhances the realism of the virtual environment and, importantly, improves the patient's learning rate and motivation.
  • Dynamic, Recovery-Based Assistance is Key: The system's ability to continuously monitor and adapt the patient's recovery level, and consequently adjust the therapist's assistance, is a critical factor in accelerating progress. This moves beyond static assistance levels.
  • Attribute-Wise Recovery Enables Targeted Training: The concept of breaking down recovery into specific attributes (amplitude, frequency, tremor) is a powerful idea. It allows for highly personalized training, addressing individual weaknesses more effectively than a single, generalized recovery metric.
  • Simplified Architecture Can Achieve Superior Performance: By proposing a more streamlined multilateral architecture, the authors demonstrate that complexity is not always synonymous with effectiveness. Their system achieves better performance with fewer channels.
  • Stability is Achievable with Full Force Feedback: A significant challenge in shared control is maintaining stability when providing full force feedback. The paper's controller design and passivity analysis show that this is achievable, even in challenging environments like a virtual wall.

Key Results

  • Improved Recovery Growth: Patients showed a continuous growth in recovery, reaching a level of 0.84 within a 1-minute session, a characteristic often missing in state-of-the-art methods.
  • Enhanced Position Tracking: The proposed architecture demonstrated superior position tracking compared to the traditional architecture, with lower RMSE values across all axes (e.g., RMSE in X reduced from 16.99 mm to 12.42 mm).
  • Faster Error Correction: In a frequency mismatch scenario, the proposed architecture was able to match the therapist's trajectory in approximately 12 seconds, compared to 22 seconds with the traditional architecture.
  • Effective Handling of Tremor and Amplitude Mismatch: The attribute-wise recovery factor allowed the proposed system to handle scenarios with minor frequency differences and tremor more effectively than the traditional approach.
  • System Stability Confirmed: The system remained stable even during collisions with a high-stiffness virtual wall, empirically confirming the effectiveness of the proposed controllers.

Limitations and Future Work

  • Limitations:
    • The study used a healthy subject behaving as a patient due to IRB approval limitations. Future work should involve testing with actual stroke patients.
  • The current implementation is for upper extremity movements. Extending it to other body parts would be beneficial.
  • While attribute-wise recovery is proposed, the experimental validation in Section 6 was primarily simulation-based. Further experimental validation with real patients is needed.
  • Future Work:
    • Conducting experiments with actual stroke patients to validate the findings.
  • Extending the architecture and controllers to other body parts for comprehensive rehabilitation.
  • Further experimental validation of the attribute-wise recovery concept with real-world patient data.
  • Exploring more advanced machine learning techniques for dynamic recovery factor estimation and adaptive assistance.

Summary updated

BibTeX
@article{annamraju2025robotic,
  author = {Srikar Annamraju and Harris Nisar and Anne Horowitz and Dusan M. Stipanovic},
  title = {{Robotic Rehabilitation Through Multilateral Shared Control Architecture}},
  journal = {Robotics},
  year = {2025},
  doi = {10.3390/robotics14040050},
}
Figure