Development of Robotic Ankle Rehabilitation System to Enhance Human Machine Interaction
- Publication Type:
- Thesis
- Issue Date:
- 2023
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Ankle injuries are quite prevalent and are one of the leading factors that might prevent a person from engaging in daily activities. These injuries may result from running, falling, a stroke or several other injuries. Regaining complete ankle functionality after severe injuries requires therapy. Rehabilitation is one of the required methods that might expedite patients' recoveries from ankle injuries. In recent years, rehabilitation has included technologies, such as robot-assisted rehabilitation. Research has shown the advantages of robot-based rehabilitation, but there are several factors that must be addressed to have a better experience while engaging with the robot.
This thesis proposes methodologies to improve human-machine interaction by enabling control of the robot via physiological signals, reducing computational, providing an adaptive control system and improving user safety. Initially, a mechanical structure for the robot which the users can easily use was designed. To control the robot via physiological signals, classifiers were used to recognise the intended movement using the user's electromyogram (EMG) and electroencephalogram (EEG) signals. A combination of features was used where classifying the EMG signal achieved a 98.9% accuracy. IMU sensors have been employed to ensure user safety during while performing the exercises. Ankle plantarflexion and dorsiflexion torque and joint angle are estimated using EMG. Once the intend has been determined, the activation functions acquired from a single EMG channel are used with nonlinear models to estimate joint torque and joint angle. Results from the experiments proved the proposed strategy to be effective.
Movement classification based on EEG enabled the user to operate the robot by imagining the desired movement. A novel channel reduction strategy using ICA was proposed where EEG channels were decreased to 50% to achieve the same accuracy as using all EEG channels. An MRMR features reduction approach was studied to lower the size of features. The findings demonstrated that employing just the dominant features to classify the signal had a negligible effect on accuracy.
After establishing the ankle joint's angle, this thesis's last module discusses a control system to adapt to the user's movement. Reinforcement learning, a self-learning machine-learning method, is utilised to train an agent to drive the robot. For this function, a deep deterministic policy gradient method was applied. The algorithm tunes a PID controller such that its parameters were changed in real-time based on the user's movement. The results demonstrate that the robot can follow the user's movement with less than 2% error.
Please use this identifier to cite or link to this item: