Various prostheses, exoskeletons, and orthoses may require an algorithm or system in place to instruct their function. Ideally, this system possesses integrative qualities that enable it to predict and/or respond to the intended activities of the user with consistent accuracy. Statistical modeling has been utilized in pursuit of this goal, but it is not entirely effective in consistently predicting human activity based upon a developed knowledge and understanding of intent and action. Therefore, there is a need for a more adaptable and accurate means for predicting human activity and allowing for computer systems to be more effectively integrated with their users.
Researchers at Arizona State University have developed an approach for the classification of human activity based upon the periodic nature of basic human activities such as jogging and walking. Because of this periodic nature, the angular frequency (ω) and amplitude (A) of the signals generated with different types of movement can be utilized in a machine learning algorithm employing a method called A – ω feature extraction, which is a method of calculating angular frequency and amplitude from the signals produced when subjects performed each activity. Machine learning, in place of statistical modeling, provides an improved ability to learn from obtained data and make predictions based on the algorithm being used.
The results of the experiments with various subjects and activities demonstrated a clear difference in A – ω features between the varying activities. The machine learning algorithms then are effective means for classifying the A – ω features and thus predicting human activity based upon the signals generated when a subject was intending to perform the activity, with near-100% accuracy.
• Prostheses, orthoses, and exoskeletons
Integrating the intent of the user more seamlessly and consistently with the activity of the machine itself
Benefits and Advantages
• Accuracy: more consistently accurate in predicting and reacting to intended activity
• Self-Learning: ability to learn from data without continuously programming instructions
• Integration: enables machinery to integrate more seamlessly with biological function