ロボット装着者の行動検出のための機械学習モデルの評価

抄録

A human body support robot has been attracted, because a robotic technology is expected to support these aging people for walking trainings or rehabilitation trainings, and workers for heavy lifting. These robots tend to mount a single assist control system, walking assist or lifting assist. In this case, the robot does not assist for human when human perform the action that does not mount an assist control system in the robot. When the robot does not assist for human, human suffers from the robot load. When the robot has a multiple assist control system, human needs to change the assist control systems, and this assistance is not seamless. The purpose of this research is to develop the function that automatically changes assist control systems by detecting a human action. This paper describes the method for detecting human action using machine learning. We build the prediction models based on decision tree. We verified a percentage of correct detection and a tree structure for mounting in the robot. In the result, the percentage of correct detection is more than 90%, and we are able to build the prediction model, which the tree structure is simple.