Intelligent Gait Mode Detection in Wearable Robots

This paper investigates machine learning approaches for gait mode detection in lower limb exoskeletons. Five gait modes (level walking, stair ascent, stair descent, ramp ascent, and ramp descent) are identified using a limited set of sensor data (IMU and heel motion markers) to ensure fast and accurate detection. The performance of 31 classification models trained on two different datasets is studied and four of the best models in terms of accuracy are demonstrated and discussed. Results show that a Bilayered Neural Network (BNN) trained on equalized data achieves the best performance with a success rate of 97.57% in real-time simulation. The BNN outperforms a similar Multi-Layer Perceptron Neural Network (MLPNN) model used in previous research, demonstrating its potential for robust gait mode detection in wearable robots.