Neural network-based hybrid human-in-the-loop control for meal assistance orthosis.

In order to assist some elderly and disabled people, who have partly or completely lost the ability of moving their upper limbs due to neurological disabilities or spinal cord disease, to take meals by themselves independently, a new type of meal assistance orthosis was recently developed. This paper presents a neural network-based hybrid human-in-the-loop control for this meal assistance orthosis with functional and safety purposes. In this approach, the position control and the force-free control are integrated into a single controller based on the model of meal assistance orthosis. By means of the position control, the meal assistance orthosis is controlled to generate appropriate compensation forces for assisting the movement of upper limb. In order to reduce the risk of hurting the bodies of human end-users and of damaging the device due to the impact from large external forces, with the force-free control, the meal assistance orthosis can flexibly move with the driven of large external forces. In addition, the controller of the meal assistance orthosis can be smoothly switched between the position control and the force-free control through a designed process to avoid instantaneously generating large external force owing to hard switching. In order to improve the adaptability of the proposed approach to different subjects, neural networks are adopted in the controller. Moreover, the proposed approach fully takes into account the influence of external forces induced by upper limb in the control process to form a kind of human-in-the-loop control. With the simulation and experiment of the meal assistance orthosis, the effectiveness of the proposed method was verified.