Reference-based Positive Ski Coaching using Foot Pressure and MLLMs

We propose a system that generates positive ski coaching by comparing foot-pressure sequences from insole sensors. Coaching plays an important role in sports learning, and positive feedback is known to increase confidence and support skill acquisition. Existing coaching systems prioritize error identification and corrective feedback, often overlooking the importance of positive feedback. Our approach uses pairs of pre- and post-improvement movements from an existing coaching dataset and converts the associated corrective feedback into positive feedback to construct a new dataset. We then generate positive coaching from two input foot-pressure sequences. Experimental results demonstrated that the reference-based model outperformed existing approaches in generating both corrective and positive feedback.