Related Experiment Video
Updated: Sep 26, 2025
Quantified Assessment of Infant's Gross Motor Abilities Using a Multisensor Wearable
Published on: May 17, 2024
Performance of Deep Learning Models in Forecasting Gait Trajectories of Children with Neurological Disorders
Rania Kolaghassi1, Mohamad Kenan Al-Hares1, Gianluca Marcelli1
1School of Engineering, University of Kent, Canterbury CT2 7NT, UK.
Insights
Deep learning models forecast children's pathological gait trajectories for robotic device control. Long Short-Term Memory (LSTM) networks outperformed Convolutional Neural Networks (CNNs), showing promise for advanced rehabilitation technologies.
Area of Science:
- Robotics
- Biomedical Engineering
- Artificial Intelligence
Background:
- Forecasting gait trajectories is crucial for controlling lower limb robotic devices like exoskeletons.
- Existing research primarily focuses on healthy gait, neglecting pathological gait patterns in children.
- Pathological gait in children, common in conditions like cerebral palsy, presents unique variability challenges for control systems.
Purpose of the Study:
- To implement and compare deep learning models (LSTM and CNN) for forecasting pathological gait trajectories in children.
- To investigate the impact of input and output time-frames on prediction accuracy.
- To establish the feasibility of using forecasted gait data for controlling rehabilitative robotic devices.
Main Methods:
- Two deep learning models, Long Short-Term Memory (LSTM) and Convolutional Neural Network (CNN), were developed.
- Models were trained on motion capture data of children (ages 4-19) with neurological disorders, including cerebral palsy.
- Gait trajectories (hip, knee, ankle Euler angles) were forecasted up to 200 ms into the future, with varying input/output window sizes.
Main Results:
- LSTMs demonstrated superior performance over CNNs, with prediction error differences increasing with larger input/output window sizes.
- Mean Absolute Errors (MAEs) ranged from 0.095-2.531 degrees for LSTM and 0.129-2.840 degrees for CNN.
- Input window size had minimal impact on errors for output windows ≤50 ms, but larger input windows reduced errors for longer output windows.
Conclusions:
- Deep learning models can successfully forecast pathological gait trajectories in children.
- LSTM networks offer higher accuracy for gait prediction compared to CNNs in this context.
- This forecasting capability holds significant potential for enhancing the control systems of pediatric rehabilitative robotic devices.
Abstract:
Forecasted gait trajectories of children could be used as feedforward input to control lower limb robotic devices, such as exoskeletons and actuated orthotic devices (e.g., Powered Ankle Foot Orthosis-PAFO). Several studies have forecasted healthy gait trajectories, but, to the best of our knowledge, none have forecasted gait trajectories of children with pathological gait yet. These exhibit higher inter- and intra-subject variability compared to typically developing gait of healthy subjects. Pathological trajectories represent the typical gait patterns that rehabilitative exoskeletons and actuated orthoses would target. In this study, we implemented two deep learning models, a Long-Term Short Memory (LSTM) and a Convolutional Neural Network (CNN), to forecast hip, knee, and ankle trajectories in terms of corresponding Euler angles in the pitch, roll, and yaw form for children with neurological disorders, up to 200 ms in the future. The deep learning models implemented in our study are trained on data (available online) from children with neurological disorders collected by Gillette Children's Speciality Healthcare over the years 1994-2017. The children's ages range from 4 to 19 years old and the majority of them had cerebral palsy (73%), while the rest were a combination of neurological, developmental, orthopaedic, and genetic disorders (27%). Data were recorded with a motion capture system (VICON) with a sampling frequency of 120 Hz while walking for 15 m. We investigated a total of 35 combinations of input and output time-frames, with window sizes for input vectors ranging from 50-1000 ms, and output vectors from 8.33-200 ms. Results show that LSTMs outperform CNNs, and the gap in performance becomes greater the larger the input and output window sizes are. The maximum difference between the Mean Absolute Errors (MAEs) of the CNN and LSTM networks was 0.91 degrees. Results also show that the input size has no significant influence on mean prediction errors when the output window is 50 ms or smaller. For output window sizes greater than 50 ms, the larger the input window, the lower the error. Overall, we obtained MAEs ranging from 0.095-2.531 degrees for the LSTM network, and from 0.129-2.840 degrees for the CNN. This study establishes the feasibility of forecasting pathological gait trajectories of children which could be integrated with exoskeleton control systems and experimentally explores the characteristics of such intelligent systems under varying input and output window time-frames.
