loading page

A Deep-Learning Based Real-Time Prediction of Seated Postural Limits and its Application in Trunk Rehabilitation
  • +3
  • Xupeng Ai ,
  • Victor Santamaria ,
  • Jiawei Chen ,
  • Boce Hu ,
  • Chenfei Zhu ,
  • Sunil Agrawal
Xupeng Ai
Columbia University in the City of New York, Columbia University in the City of New York

Corresponding Author:[email protected]

Author Profile
Victor Santamaria
Author Profile
Jiawei Chen
Author Profile
Chenfei Zhu
Author Profile
Sunil Agrawal
Author Profile


Seated postural limit defines the region from beyond which a subject cannot return the trunk to the neutral position without additional external support. The seated postural limits can be used as a reference to provide assistive support to the torso by an appropriately designed robotics, e.g., Trunk Support Trainer (TruST). However, fixed boundary representations of seated postural limits in the literature cannot capture the dynamically changing seated postural limits during training. In this study, we propose a conceptual model of the dynamic boundary of the trunk center using a boundary vector designed to track the postural-goal direction and trunk movement amplitude during a sitting task. We experimented with 20 healthy subjects. The results support our hypothesis that TruST intervention with an assist-as-needed controller based on the proposed dynamic boundary representation could achieve more significant sitting workspace area improvements than a fixed boundary representation. The second contribution of this paper is that we provide an approach to effectively introduce deep learning into TruST’s real-time controller design. We compiled a 3D trunk movement dataset which is currently the largest in the literature. We designed a loss capable of solving the gate-controlled regression problem. We proposed a novel roadmap for the exploration study. Following the roadmap, we developed a deep learning architecture, modified the widely used Inception module, and then obtained a deep learning model capable of accurately predicting the dynamic boundary in real-time. We believe this approach can be extended across other rehabilitation robots towards designing intelligent dynamic boundary-based assist-as-needed controllers.
2023Published in IEEE Transactions on Neural Systems and Rehabilitation Engineering volume 31 on pages 260-270. 10.1109/TNSRE.2022.3221308