I Hsum Li

This paper proposed a novel design using a torsion spring mechanism with a single fluidic muscle actuator (FMA) to drive a joint with one degree-of-freedom (DOF) through a steel wire and a proportional pressure regulating valve (PRV). We developed a 4-DOF wearable upper-limb rehabilitation assistance exoskeleton system (WURAES) that is suitable for assisting in the rehabilitation of patients with upper-limb injuries. This system is safe, has a simple mechanism, and exhibits upper-limb motion compliance. The developed WURAES enables patients with upper-limb musculoskeletal injuries and neurological disorders to engage in rehabilitation exercises. Controlling the joint is difficult because of the time-varying hysteresis properties of the FMA and the nonlinear motion between standard extension and flexion. To solve this problem, a proxy-based output feedback sliding mode control (POFSC) was developed to provide appropriate rehabilitation assistance power for the upper-limb exoskeleton and to maintain smooth and safe contact between the WURAES and the patient. The POFSC enables the overdamped dynamic of the WURAES to recover motion to be aligned with the target trajectory without a significant error overshoot caused by actuator saturation. The experimental results indicate that the proposed POFSC can control the designed WURAES effectively. The POFSC can monitor the exoskeleton system’s total disturbance and unknown state online and adapt to the exterior environment to enhance the control capability of the designed system. The results indicate that a single FMA with a torsion spring module exhibits a control response similar to a dual FMA configuration.

The development of autonomous mobile robots (AMR) has brought with its requirements for intelligence and safety. Human action recognition (HAR) within AMR has become increasingly important because it provides interactive cognition between human and AMR. This study presents a full architecture for Edge-AI human action recognition (Edge-AI HAR) to allow AMR to detect human actions in real time. The architecture consists of three parts: a human detection & tracking network, a key frame extraction function, and a human action recognition network. The human action recognition network is a cascade of a DenseNet121 and a two-layer bi-directional LSTM (DLBiLSTM), in which the DenseNet121 is a pre-trained model to extract spatial features from action key frames and the DLBiLSTM provides a deep two-directional LSTM inference to classify complicated time-series human actions. Edge-AI HAR undergoes two optimizations - ROS distributed computation and TensorRT structure optimization - to give a small model structure and high computational efficiency. Edge-AI HAR is demonstrated in two experiments using an autonomous mobile robot and is demonstrated to give an average precision of 97.58% for single action recognition and around 86% for continuous action recognition.