Detection of Abnormal Signs through Analysis of Grip Failure Data

Against the labor shortage in the restaurant business, automatic dishwashing robots have been being developed. Such the present robots have difficulty of grasping dishes, which causes failure by abnormal grasping. Thus, this paper proposes a method to detect a sign of abnormity from a grasping action through an acceleration sensor attached to the robot. The proposed method first applies the sliding window for each time point in the sensor time-series data and obtains plural time-subseries. By SVD of the matrix generated by these subseries, the variation of the sensor value is expressed as a linear subspace. The dissimilarity degree for each time point is calculated from the distance between the corresponding subspaces among two time-series data. By comparing the average dissimilarity between each normal data and test data with the average dissimilarity degree among normal data, the sigh of abnormity in the test data can be detected.Theexperimental result with an actual robot showed that the proposed method enabled detection of the sign of abnormity. It also showed that the cause of abnormity can be distinguished from the term detected as the sign of abnormity during the grasping action.