Motor Control of the Baseball-Hitting Technique using Artificial Intelligence Feedback
(Donghwi Suh, Young-Vin Kim)

This study investigated whether machine learning could classify baseball-hitting techniques and whether artificial intelligence feedback could improve the motor task outcome and movement coordination. For the machine learning model, 48 skilled athletes and 440 novices were recruited. A support vector machine was used as a machine learning method for skill classification. For the pre-test, practice, post-test, and retention tests in motor learning, additional 42 general adolescents were recruited, aside from the already included 488 participants. Motor learning participants were randomly divided into three groups: a control group (N = 14), an artificial intelligence feedback (AIfb) group (N = 14), and a general feedback (Gfb) group (N = 14). Machine learning successfully classified the skill level of baseball hitting. The absolute error of the launch angle was smallest in the Gfb group during the post-test, while the variable error showed the greatest decrease in the AIfb group. The exit velocity was higher in the AIfb than in the Gfb group after practice. The AIfb group exhibited the highest ratio of sequential accelerations among the pelvis-torso-left elbow joints. The number of principal components decreased to three and two in the AIfb and Gfb groups, respectively. As for the loading data, AIfb commonly included the upper limbs, pelvis, and the lower limbs in the group with coordinated movements; for Gfb, after motor learning, the coordination pattern of the lower limbs was eliminated, and the coordination pattern involving only the upper limbs was also removed after practice. Consequently AIfb contributed to improved movement coordination and could effectively change the dynamical degrees of freedom. This technology could benefit coaching in a real-life setting and aid in understanding the human movement coordination structure.