Keisuke Fujii - Research

Summary of my research

When I was a Ph. D. student and post-doc, my research areas were sports science (e.g., motion analysis in basketball), human behavior (e.g., locomotion), and robotics. Recently, due to the development of measurement technology, it has become possible to measure complex motion data of actual sports games. Therefore, I currently utilize frameworks in machine learning, dynamical systems, signal processing, and other areas such as computational biology for such complex motion data. My current main research topics are summarized as follows

· Classification of (collective) motion patterns

· Prediction of future motions

· Control for performing skilled movements

Especially, my interests are skillful movements and teamwork such as in basketball and football (additionally, I am also interested in my collaborator's topics such as golf [18], children's play [24], locomotion [e.g., 19], and standing [e.g., 21]).

Classification of (collective) motion patterns We develop computational methods of automatically classifying complex (collective) motion patterns such as in sports for the understanding and evaluation of the motions. It will contribute to reducing the workload of people who evaluate the motion by watching videos for a long time. For example, we studied strategic cooperative plays in sports [27], a resilient team-helping against the cooperative attacks [16], and more general classification methods in collective motion [26].

Prediction of future motions From complex multivariate data such as sports, we develop computational methods to predict or explain the final outcome or objective of the motion (e.g., winning or score). Our study will reveal the critical movements contributing to the outcome without domain knowledge. For example, we studied a critical kinetic factor in 1-vs-1 outcome [11, 10], a prediction method of movement direction from posture [7] and score in basketball [26].

Control for performing skilled movements After revealing the characteristics of skillful movements by the above approaches, the realization of such movements is needed. Humans can skillfully control a great number of components (e.g., muscles and joints), but there is still unknown about flexible and dynamic collective motion. If this movement can be computationally realized, there is a possibility to clarify a part of the mechanism. We can also verify the effectiveness of the characteristics of the motion clarified by the above approaches, and simulate the unknown movement which is difficult to execute in a real world. For example, we modeled a 1-vs-1 cognitive-motor system [13], a switching autonomous motor control system [20]. In the future, we want to model a multi-agent model to understand teamwork (e.g., [23]).