Related Experiment Video
Updated: Oct 11, 2026

Using a Split-belt Treadmill to Evaluate Generalization of Human Locomotor Adaptation
Published on: August 23, 2017
Multimodal adaptive mirror therapy through gait adaptation on a split-belt treadmill for gait training
Siyu Yang1, Kangjie Zheng1, Cenwei Li1
1School of Automation and Intelligent Manufacturing, Southern University of Science and Technology, Shenzhen, Guangdong, China.
Abstract:
BackgroundTraditional mirror therapy relies on fixed kinematic symmetry, often neglecting the trade-offs between movement accuracy, stability, and muscular effort. This study developed a Multimodal Adaptive Mirror Therapy system featuring a multi-objective controller for personalized gait rehabilitation.ObjectiveTo develop and evaluate a multimodal adaptive mirror therapy system that integrates gait kinematics, postural stability, and muscular activation into a unified multi-objective optimization framework for personalized gait rehabilitation.MethodsTwenty-six subjects underwent asymmetric gait training with unilateral limb restraint to simulate asymmetric gait patterns. The proposed system integrated camera-based gait data, inertial measurement unit (IMU)-based stability metrics, and electromyography (EMG) activation into a unified objective function. Optimization was dynamically adjusted via a weighted harmonic mean of multimodal scores. Normalized mutual information (NMI) was used to quantify dependencies, while interquartile range (IQR) analysis evaluated controller stability.ResultsNMI analysis revealed significant dependencies between all modalities (p < 0.05), with the training parameter showing the strongest correlation with stability (NMI = 0.342). The multimodal controller achieved balanced, near-optimal performance, yielding median scores of 97.81% for stability and 92.44% for gait. Notably, it exhibited the lowest inter-participant variability among all strategies in the stability domain (IQR: 24.80%), compared to substantially wider dispersions observed under unimodal controllers.ConclusionThis study demonstrated that a multimodal, closed-loop control scheme can effectively resolve performance trade-offs in gait training. By transitioning from rigid imitation to dynamic multi-objective optimization, the system provided a robust, intelligent closed-loop platform, establishing a practical foundation for patient-centered robotic rehabilitation.

