Marina Aoyama

Programming by demonstration is an effective way of teaching robots to learn tasks from demonstration and perform in an unseen situation. It is widely used in robotics as it enables faster and more flexible learning compared to manual programming and it does not require programming skills of the teacher. There is extensive research on developing efficient imitation learning algorithms, however, there are only a few studies on how to provide the optimal demonstration although the performance is highly dependent on the quality of teaching data. Therefore, my research project aims to derive the optimal teaching strategy for teachers to produce good quality demonstration which allows robots to achieve high performance with a low teaching cost. Finding such a teaching strategy is called Machine Teaching and it plays an especially important role when teaching robots because the teaching cost tends to be high and good quality data is essential to reduce the cost.