Abstract Summary/Description
The mridangam is the principal percussion instrument in Carnatic (one form of Indian classical) music and demands a rare combination of finger dexterity, continuously variable stroke dynamics, and precise timbral control to properly articulate rhythmic syllables (Solkattu). Transferring and assessing these highly nuanced, culturally embedded motor skills has remained a significant challenge for both traditional pedagogy and modern technological approaches. This paper presents the Robotic Mridangam Guru, a human-robot interaction system developed to teach Carnatic percussion and preserve its technical and expressive depth. The system consists of two tightly integrated components: 1. An autonomous intelligent tutor that acts as a practice partner by providing real-time, multimodal performance evaluation. 2. An expressive robotic demonstrator equipped with biomimetic end-effectors capable of reproducing the full range of mridangam strokes and timbres with near-human fidelity. The tutoring module combines state-of-the-art computer vision for accurate hand and finger tracking with deep audio models trained to recognize Solkattu patterns and evaluate timbral quality. By fusing visual kinematics with detailed acoustic analysis, the system assesses both the geometric precision of strokes and their sonic outcome. Feedback is delivered instantly through an interactive interface, enabling effective independent practice without requiring a human teacher. Beyond Carnatic music, the framework offers a scalable approach for documenting, transmitting, and revitalizing endangered performative traditions worldwide through embodied robotic tutors.