Motion Intent Recognition Algorithm for Lower Limb Rehabilitation Robots Based on Dual-Physics Sensor Feature Fusion

Abstract

Existing motion intent recognition systems in lower limb rehabilitation robots primarily rely on the fusion of multiple sensor features. Such systems capture the motion characteristics of healthy volunteers during specific movements and then process the data using machine learning algorithms to accurately recognize human motion events, such as forward and backward movements. We address the complexity and inaccuracies of current intent recognition systems by synthesizing feedback from rehabilitation physicians and patients and adopting modular design principles to develop an integrated human motion intent recognition system for lower limb rehabilitation robots. The system utilizes dual physical sensors to collect data on the movement characteristics of the patient’s waist, abdomen, and shoulders, which are then classified using the Transformer-LSTM algorithm. The dataset employed for training and testing the algorithm was gathered from a tertiary care hospital, focusing on the movement characteristics of patients with functional hemiplegia of the lower extremities. Clinical trial results demonstrated that the Transformer-LSTM algorithm achieved an average classification accuracy of 97.54% in recognizing human lower limb movement events, compared to 87.48% with the LSTM algorithm. This lower limb rehabilitation robot also significantly enhances patient motivation and comfort during training.