In a significant stride towards enhancing the functionality of active prosthetics, researchers have developed a deep learning model that could revolutionize the way prosthetics mimic natural human movement. The study, led by Mohamed Karakish from the Mechanical Engineering Department at the Arab Academy for Science, Technology and Maritime Transport (AASTMT) in Cairo, was recently published in the journal ‘Discover Applied Sciences’ (translated to English as ‘Discover Applied Sciences’).
The research tackles the complex challenge of replicating natural human gait, which is inherently non-linear and intricate. The team designed a three-step deep learning model that processes kinematic data from the shank—linear acceleration and angular velocity—to classify activities, identify gait phases, and generate ankle trajectories. This sequential approach aims to improve the control systems of active prosthetics, making them more responsive and natural.
The study compared two deep learning configurations: Multilayer Perceptrons (MLP) and Convolutional Neural Networks (CNN). Both models achieved high accuracy in differentiating between static and dynamic activities, with MLP offering significantly faster performance on embedded microcontrollers like the ESP32/ESP32S3. “MLP could achieve very close accuracy in comparison to CNN while having much faster performance on the embedded microcontroller,” noted Karakish.
For static activities, MLP achieved an inference time of 3.154 ms, approximately 86.2% faster than CNN. In dynamic activities, MLP was even more efficient, with an inference time of 4.247 ms compared to CNN’s 83.032 ms. The study also found that MLP was about 97% faster than CNN for walking and running trajectories, with a slightly lower root mean square error (RMSE). However, for bicycling, CNN delivered more accurate trajectories despite a higher RMSE.
The implications of this research extend beyond medical prosthetics. In the maritime sector, where human activity recognition and gait analysis could enhance safety and efficiency, these models offer promising opportunities. For instance, monitoring the gait of crew members could help in early detection of fatigue or injury, reducing the risk of accidents. Additionally, the real-time trajectory generation could be adapted for robotic systems used in maritime operations, improving their responsiveness and precision.
The study highlights the potential of deep learning in creating more intuitive and responsive control systems for prosthetics and other applications. As Karakish explained, “when sufficient phase information is available, the MLP can achieve accuracy comparable to the CNN while offering significantly faster prediction times.” This balance between accuracy and speed is crucial for real-time applications, making the research particularly relevant for industries requiring quick and precise data processing.
In summary, the research published in ‘Discover Applied Sciences’ presents a significant advancement in the field of human activity recognition and gait analysis. The findings offer valuable insights for the maritime sector, where enhancing safety and efficiency through advanced technology is a priority. As the technology matures, it could lead to innovative solutions that improve the quality of life for individuals with prosthetics and enhance operational efficiency in various industries.