An Upper-Limb Power-Assist Exoskeleton Using Proportional Myoelectric Control

Kejun Zhang(张克俊), Zhichuan Tang, Kejun Zhang, Shouqian Sun, Zenggui Gao, Lekai Zhang, Zhongliang YangSensors 2014 · DOI

TL;DR

This paper presents an upper-limb exoskeleton that uses pneumatic muscles for power assistance, controlled by proportional myoelectric signals derived from EMG. The system successfully translates user's intended motion into real-time exoskeleton control, demonstrating its potential for augmenting arm performance and aiding in rehabilitation after neurological injury.

What Problem Does This Paper Address?

Assisting individuals with physical disabilities or those undergoing rehabilitation often requires robotic devices that can augment their natural movements. Upper-limb exoskeletons are particularly challenging due to the complexity and dexterity required for everyday tasks. While electromyography (EMG) signals offer a direct way to infer user intent by measuring muscle activity, traditional "on-off" control methods lack the nuance for smooth, proportional assistance. This paper addresses the need for a more intuitive and responsive control system for upper-limb exoskeletons, enabling them to provide effective power assistance that aligns with the user's real-time motion intentions.

Methods

The core of this research lies in developing a proportional myoelectric control scheme for an upper-limb exoskeleton. The exoskeleton itself is actuated by pneumatic muscles, chosen for their safety and compliance. The control system involves several key steps:

  • EMG Signal Acquisition: Surface EMG signals are collected from four key muscles in the arm: biceps brachii, brachioradialis, triceps brachii, and anconeus.
  • Feature Extraction: The raw EMG signals are processed through preamplification, analog-to-digital conversion, and filtering. To represent the muscle activity, the Root Mean Square (RMS) of the filtered EMG signals is calculated as a time-domain feature.
  • EMG-Angle Model Construction: A Back-Propagation Neural Network (BPN) is employed as a classifier. This three-layer BPN takes the RMS values from the four muscles as input and is trained to predict the elbow joint angle. This creates an "EMG-angle model" that maps muscle activity to intended joint movement.
  • Proportional Control: The predicted elbow angle is then used to generate control signals for the pneumatic muscles. This is achieved through a series of equations that translate the desired angle into the necessary air pressure for the actuators, enabling proportional control rather than simple on-off activation.

Datasets

The study utilized data collected from six healthy male subjects. For training and testing the EMG-angle model, data was collected during specific experimental protocols. For each subject, sixty sets of movement data were collected under different motion periods (2-s, 4-s, and 8-s). Forty-eight sets were randomly selected for training the BPN, and the remaining twelve were used for testing. No pre-existing public datasets were used; the data was generated specifically for this study.

Key Insights and Ideas

  • Proportional Control is Key: The research strongly emphasizes that proportional myoelectric control, as opposed to on-off control, is crucial for achieving smooth and intuitive assistance from exoskeletons. This allows the exoskeleton to better match the user's intended movement speed and force.
  • EMG-Angle Mapping is Feasible: The successful development of an EMG-angle model using a BPN demonstrates that it's possible to reliably predict a user's intended joint angle from their muscle activity. This mapping is adaptable to individual subjects through the training process.
  • Motion Period Matters for Prediction: The study found that the duration of the motion period significantly impacts the prediction performance of the EMG-angle model. The four-second period yielded the best results, suggesting an optimal balance between capturing sufficient muscle activity and avoiding issues related to rapid acceleration/deceleration or prolonged isometric contractions.
  • Pneumatic Muscles Offer Safe Actuation: The choice of pneumatic muscles for actuation highlights their suitability for human-robot interaction, offering inherent compliance and safety compared to more rigid actuators.
  • Power-Assist Effectiveness Demonstrated: The experimental results clearly show a reduction in muscle activity (EMG and RMS) when the exoskeleton is actively assisting the user, indicating that the device is effectively augmenting the user's effort.

Key Results

  • Best Prediction Performance: The EMG-angle model achieved the best prediction performance under the four-second motion period, with a Root Mean Square Error (RMSE) of 9.67 and an R² value of 0.87. This was superior to the two-second period (RMSE: 10.70, R²: 0.83) and the eight-second period (RMSE: 12.42, R²: 0.79).
  • Power-Assist Effectiveness: Figures 9 and 10 show a clear reduction in EMG amplitude and RMS values for the four muscles when the exoskeleton was actively controlled (Experiment Four) compared to when it was worn with no actuation (Experiment Three) or not worn at all (Experiment Two). This reduction signifies that the exoskeleton was successfully providing power assistance, reducing the user's muscular effort.
  • Real-time Control Achieved: The system demonstrated the ability to control the exoskeleton in real-time based on the user's motion intention, allowing for smooth and adaptable assistance.

Limitations and Future Work

  • Muscle Fatigue: The study acknowledges that muscle fatigue, potentially induced by the repetitive nature of the experiments (over 200 movements in some cases), could affect EMG signals and thus the prediction performance. This is identified as an important factor for future investigation.
  • Pneumatic Muscle Limitations: The inherent limitations of pneumatic muscles, such as their limited contraction range (about one-third of initial length), restricted the achievable range of motion for the elbow joint. The authors reduced the joint diameter to achieve a 0°-90° range, which is still less than typical human anthropometric boundaries.
  • Portability and Flexibility: The heavy air supply equipment (air pump) limited the portability and flexibility of the exoskeleton, preventing users from moving around freely while wearing it.
  • Individual Differences: A significant challenge is the variability in EMG signals even from the same motion, even within the same person. This necessitates system training (BPN) for each subject to adapt to their unique signal patterns.
  • Future Directions:
    • Investigating the impact of motion acceleration and speed on prediction performance.
  • Exploring alternative actuators like servo motors or hydraulic cylinders to overcome the limitations of pneumatic muscles and improve portability.
  • Developing more robust EMG electrode technology and optimizing control algorithms to address individual differences.
  • Considering sensor fusion approaches (combining EMG with other sensor modalities) for more complex control tasks and improved clinical viability, especially for controlling multiple degrees of freedom.

Summary updated

BibTeX
@article{zhang2014upper,
  author = {Kejun Zhang(张克俊) and Zhichuan Tang and Kejun Zhang and Shouqian Sun and Zenggui Gao and Lekai Zhang and Zhongliang Yang},
  title = {{An Upper-Limb Power-Assist Exoskeleton Using Proportional Myoelectric Control}},
  journal = {Sensors},
  year = {2014},
  doi = {10.3390/s140406677},
}
Figure