Personalized gait trajectory generation for lower-limb rehabilitation robots using anthropometric features

Conventional gait generation schemes fail to accommodate variable walking speeds and diverse rehabilitation scenarios. This study constructs an improved Gaussian process regression (IGPR) model with hybrid kernel functions based on 12 anthropometric indicators, walking speed and training modes. The dataset consists of 42 healthy participants, with averaged hip joint angle trajectories RMSE = 3.31° and knee joint angle trajectories RMSE = 4.37°. We separate model prediction, offline gait generation and robot tracking verification. Tracking tests on four healthy volunteers prove stable trajectory following. This method supplies quantitative gait references for wearable devices, further clinical trials with stroke patients are required to verify rehabilitation outcomes.