Personalized Rehabilitation Assistive Device Parameter Design by Integrating Human Motion-Capture Data and Artificial Intelligence

Abstract

Personalized rehabilitation assistive devices need to set parameters based on an individual's joint mobility ability, movement performance, and compensatory behaviors. Human motion capture data can provide objective kinematic information, but current artificial intelligence methods are mostly used for motion recognition and motion quality assessment, and are less used for determining device parameters. To address this issue, this study proposes PMAD-Net, a personalized motion assistance design network that combines three-dimensional skeletal trajectories and inertial measurement data. The network first aligns the two types of data onto the same time axis, then uses a spatiotemporal encoder to analyze the coordination relationships between joints, and combines individual movement characteristics for fusion. The model can be set specifically according to the movement conditions of the subjects, including the range of motion of the shoulders and elbows, auxiliary gain, virtual stiffness, the starting time and duration of the movement. The predicted results also need to meet the biomechanical constraints to avoid unreasonable settings. PMAD-Net can adjust the device settings based on the subject's movement performance, covering shoulder and elbow range of motion, auxiliary gain, virtual stiffness, as well as the start and duration of the movement. To ensure that the settings are in line with the actual movement conditions, the model's output will also undergo biomechanical constraint checks. The experiment was conducted based on the public rehabilitation movement dataset, and ten independent subjects were selected for testing. The parameters of PMAD-Net have an MAE of 0.078 ± 0.006, a motion tracking RMSE of 5.73 ± 0.39, and a constraint violation rate of 2.9 ± 0.5%. Compared to the time transformer, these three indicators have decreased by 14.3%, 10.6%, and 39.6% respectively. This method can provide equipment setting suggestions for professionals and assist in the personalized adjustment of rehabilitation equipment.