Accident Prediction Based on Motion Data for Perception-Assist with a Power-Assist Robot

Power-assist robots are expected to help facilitate the daily living motion of physically weak persons. Perception-assist has been studied to secure the safety of robot users whose sensory abilities are deteriorated or limited. The interaction between the human and environment must be carefully observed by the robot to determine the possibility of an accident in perception-assist. When the robot detects a potential accident during this interaction, it tries to avoid the accident by modifying the user's motion automatically using perception-assist. Therefore, it is important for the robot to predict potential accidents, such as falling, as soon as possible. In this paper, an accident prediction method for lower-limb perception-assist is proposed and evaluated for effectiveness. In the proposed method, the possibility of accident is predicted based on the lower-limb motion and zero-moment point of the robot user as well as information from the surrounding environment. A multilayer artificial neural network is applied in the proposed method.