外骨骼“读心术”来了！意念增强肌肉力量，深度学习预测准确率96.2%

Core Viewpoint - A new "mind-reading" exoskeleton system developed by a research team from Georgia Tech predicts user intentions 500-550 milliseconds before actions occur, providing timely assistance for individuals with limited mobility, such as stroke patients and the elderly [1][3]. Group 1: Technology and Innovation - The exoskeleton utilizes flexible bioelectronic sensors to capture muscle signals, combined with cloud-based deep learning algorithms for real-time analysis of user intentions [1][4]. - A breakthrough in sensor technology involves ultra-thin flexible EMG sensors that can be applied like a band-aid, maintaining performance even under significant strain [4][6]. - The system employs a CNN+LSTM hybrid architecture for processing EMG signals from four muscle groups, achieving classification accuracy rates of 95.38% for biceps/triceps and 97.01% for deltoid/lats [6][9]. Group 2: Performance and Effectiveness - The soft pneumatic artificial muscles (PAM) provide up to 897 Newtons of assistance, with a lightweight design that enhances the system's usability [7][9]. - The entire exoskeleton system weighs only 4.7 kilograms, including components like PAMs and batteries, and features a scalable design to accommodate users of various heights [9][10]. - Testing shows that muscle activity is reduced by 3.9 times for biceps and 3.5 times for deltoids when using the exoskeleton, even under load conditions [10][12]. Group 3: Applications and Future Development - The system supports four primary upper limb movements, effectively assisting in daily tasks such as lifting objects and using tools [9][12]. - The technology aims to enhance the quality of life for stroke patients and the elderly, with potential applications in industrial and healthcare settings [12][13]. - Ongoing development focuses on creating a universal deep learning model that can adapt to various users without individual training, leveraging cloud computing for continuous improvement [13].