Modeling the Multi-modal Behaviors of a Virtual Instructor in Tutoring Ballroom Dance

Abstract

In learning sophisticated physical skills such as sport, gymnastics, or dance, it requires the learner to practice repeatedly for long period. Embodied conversational agents (ECAs) are thus suitable candidates for such tasks. The agent (virtual instructor) can be available at anytime and suffers much less constraints in the location where to set up. The learner can thus practice the tasks for unlimited times at their favorite place at their favorite time. If the knowledge and the interaction scenario of the agent are well designed, the lessons at acceptable level can be guaranteed constantly. Unlike human instructors, they never feel tired caused by long-time service and never loose patience on awkward learners. They also do not need to mind requesting the instructor to repeat the exemplary motion and never feel embarrassed to practice when they are still unskillful. On the other hand, ballroom dance is becoming a world-wide popular sport. The motion of ballroom dance involve the whole body of the dancers, head, torso, arm, and legs have to move simultaneously and synchronized with each other and music. Ballroom dance motion are so sophisticated and require man-to-man instruction and massive repeated practice for long time. This characteristics of ballroom dance make it an appropriate candidate for research on virtual instructors.