Finding Features of Actions Efficiently Synchronized with Dishwashing Robot
In this study, we propose a method for extracting the characteristics of body motions that contribute to reducing the takt time in a cooperative task between a dishwashing robot and a human operator. The proposed method collects the takt time and motion data from novice operators until they become experienced using an inexpensive acceleration sensor. The operation data is classified into experienced and novice periods using the variance value of the takt time. In addition, the Hidden Markov Model is generated to classify the motion data into multiple motion phases. The motion features of the operator are extracted for each phase from the generated model. The proposed method finds the motion features whose difference between the experienced and novice periods are similar to the takt time transition. It uses them as important variables. We verified the effectiveness of the proposed method by conducting experiments that simulate actual work at a restaurant. The Hidden Markov Model classified the operation phases into three categories with the AUC of 0.9. In all samples, we were able to extract the motion characteristics of the experienced operators. This study showed the potential to improve the speed of novice's progress by the extracted motion characteristics to improve education guidelines and to show operators how they should physically move.