Development of a mobile gait trainer using pose estimation

Patients with neurological disorders often have gait impairments. This work aimed to develop a computer vision-based mobile gait trainer which supports users during free overground walking. The mobile gait trainer consists of three submodules: a Mecanum-wheel-driven frame with a body-weight-support (BWS) mechanism, a pose estimation module and a motor control system. Secured by the BWS, the user walked overground at preferred speeds. The webcam on the top frame estimated the shoulder movement based on the MMPose algorithms. Kinematic analysis yielded the target speeds for the trainer to follow the user. The motor control algorithms enabled the BWS to relieve the target load and the trainer to move. Preliminary test showed that the BWS mechanism produced a mean force control error of 2.03% for free walking, and 4.57% during obstacle climbing. The trainer followed the user with a speed error of 0.17 m/s. It was concluded that the trainer managed to support free overground walking.