Knee exoskeleton enhanced with artificial intelligence to provide assistance-as-needed
| Autor: | Mingxing Lyu, Wei-Hai Chen, Xilun Ding, Jianhua Wang |
|---|---|
| Rok vydání: | 2019 |
| Předmět: |
Adult
Male Knee Joint Computer science medicine.medical_treatment Electromyography 01 natural sciences Human–robot interaction 010305 fluids & plasmas Machine Learning Young Adult Artificial Intelligence 0103 physical sciences medicine Humans Exoskeleton Device Instrumentation 010302 applied physics Rehabilitation medicine.diagnostic_test business.industry Muscles Robotics Modular design Exoskeleton Trajectory Robot Female Artificial intelligence business human activities |
| Zdroj: | Review of Scientific Instruments. 90:094101 |
| ISSN: | 1089-7623 0034-6748 |
| DOI: | 10.1063/1.5091660 |
| Popis: | Robotic therapy is a useful method applied during rehabilitation of stroke patients (to regain motor functions). To ensure active participation of the patient, assistance-as-needed is provided during robotic training. However, most existing studies are based on a predetermined desired trajectory, which significantly limits the use of this method for more complex scenarios. In this paper, artificial intelligence (AI) agents are introduced to enhance the robot so that a knee exoskeleton can be autonomously controlled. A new assist-as-needed (AAN) method is proposed, where the subjects and agents cooperatively control movements. An electromyographic (EMG)-controlled knee exoskeleton with an interesting screen game is developed. Two different AI agents, modular pipeline and deep Q-network, are introduced; both can control the exoskeleton to play the screen game independently. The human-robot cooperative control is studied with two different assistant strategies, i.e., fixed assistant ratio and AAN. Eight healthy subjects participated in the initial experiment, and four assistant modes were studied. The game scores obtained by the two agents were significantly higher than those obtained by healthy subjects (EMG control), indicating that using the agents to assist stroke rehabilitation is possible. The AAN method demonstrated a better performance than the fixed assistant ratio method, indicated by the higher integral muscle activation level and participant score. Compared to a fully active control (EMG control) and fully fixed guidance (AI control), human-robot cooperative control had significantly higher integral muscle activation levels, i.e., the subjects were more involved and motivated during training. Using AI agents to power rehabilitation robots is a promising way to realize AAN rehabilitation. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|