A Practice Support System for Stepwise Learning of Stretched Backward Somersaults

Abstract

We propose a practice support system integrating step-by-step learning assistance and verbal feedback for the stretched backward somersault in artistic gymnastics floor exercises. The proposed system captures performance videos using smartphones or tablets, estimates posture from the footage, and visualizes the obtained posture information as animation-based deduction risks based on scoring rules. This feature enables users to directly understand how angle changes relate to performance evaluation. Furthermore, it analyzes performances from beginners to advanced levels, creating and displaying similarity maps between routines. Learners can select their next target performance from the map as a practice reference, enabling step-by-step learning. Additionally, it utilizes a large language model (LLM) to analyze differences between the target performance and the learner's current performance, providing improvement guidance in natural language. Evaluation experiments conducted with experienced gymnasts at beginner and intermediate skill levels confirmed the system's effectiveness in clarifying areas for improvement and enhancing learning motivation.