A Neural Network-Based Gait Phase Classification Method Using Sensors Equipped on Lower Limb Exoskeleton Robots.

Area of Science:

• Robotics
• Biomechanics
• Artificial Intelligence

Background:

• Accurate gait phase classification is crucial for controlling lower limb exoskeleton robots and understanding user intentions.
• Current methods often rely on foot sensors, which can be limiting.
• Developing advanced classification techniques is essential for enhancing exoskeleton functionality.

Purpose of the Study:

• To propose and evaluate a novel gait phase classification method for lower limb exoskeleton robots.
• To investigate the use of neural networks, specifically multilayer perceptron and NARX, for gait phase detection.
• To compare the performance of these methods against traditional foot sensors.

Main Methods:

• Utilized sensor signals from lower limb exoskeleton robots, including segment orientation and joint angular velocities.
• Developed and trained two neural network models: a multilayer perceptron and a nonlinear autoregressive with external inputs (NARX) network.
• Conducted offline and online evaluations using four distinct performance criteria.

Main Results:

• The NARX-based method demonstrated high performance in classifying gait phases.
• The proposed NARX approach achieved results comparable to or better than traditional foot sensors.
• The neural network models successfully learned to map sensor inputs to gait phase outputs.

Conclusions:

• The NARX-based gait phase classification method is a viable and effective alternative to traditional foot sensors.
• This advancement can lead to more sophisticated control and intention detection in lower limb exoskeleton robots.
• The study highlights the potential of neural networks in improving human-robot interaction within wearable robotics.