Interlimb and Intralimb Synergy Modeling for Lower Limb Assistive Devices: Modeling Methods and Feature Selection

Area of Science:

• Biomechanics and Robotics
• Human-Machine Interaction
• Machine Learning in Rehabilitation

Background:

• Gait synergy is crucial for controlling lower limb assistive devices like prostheses and exoskeletons.
• Accurate gait synergy modeling enhances human-machine interfaces by predicting movement trajectories from sound limb data.
• Optimal feature selection is vital for improving gait synergy modeling but often overlooked.

Purpose of the Study:

• To investigate optimal modeling methods and feature selection for gait synergy.
• To compare the performance of Sequence-to-Sequence (Seq2Seq), Long Short-Term Memory (LSTM), Recurrent Neural Network (RNN), and Gated Recurrent Unit (GRU) models.
• To evaluate the impact of feature selection on modeling accuracy.

Main Methods:

• Employed four neural network architectures: Seq2Seq, LSTM, RNN, and GRU.
• Utilized three feature selection methods: random forest, information gain, and Pearson correlation.
• Evaluated models for both interlimb and intralimb gait synergy prediction.

Main Results:

• Seq2Seq demonstrated superior performance over LSTM, RNN, and GRU in gait synergy modeling (MAE: 0.404° and 0.596°).
• Feature selection significantly improved Seq2Seq model performance (P < 0.05).
• The proposed Feature Selection-Seq2Seq (FS-Seq2Seq) strategy outperformed existing methods.

Conclusions:

• FS-Seq2Seq is proposed as an effective two-stage strategy for gait synergy modeling.
• This approach enables the development of synergic and user-adaptive trajectories for lower limb assistive devices.
• Improved modeling enhances human-machine interactions in assistive technology applications.