距離情報を用いた下肢外骨格型パワーアシストロボットの認知アシストにおけるリアルタイム危険推定に関する研究

抄録

To assist the elderly safely, the perception-assist technique, which can automatically change a dangerous motion to a safe motion according to the user's motion intentions and environmental conditions, has been considerably studied. In the perception-assist, detecting the environment around the user is necessary. In this study, we proposed a new obstacle segmentation algorithm based on 3D data to accurately obtain important features of obstacles in three-dimensional in real time. Based on prior research, an artificial neural network, which uses obstacles parameters, ZMP, EMG signals and motion data from the user as inputs, is applied to predict dangerous motion of the user. The effectiveness of the accident prediction, which uses the proposed obstacle segmentation algorithm, was evaluated with a lower-limb exoskeleton robot. The accuracy of accident prediction by using the obstacle parameters was 99%.