AI Can Now Watch and Grade Your Bharatanatyam Dance Moves

A team of researchers in India has taught a computer to watch Bharatanatyam — one of the world's oldest classical dance styles — and tell whether a dancer got the steps right, in the right order. The dance is built on "Adavus," 15 core step patterns with 58 variations. Each variation is a fixed sequence of poses and movements. Order matters as much as the moves themselves, which is exactly what makes it hard for a machine to judge.

Their system works in stages. First, a computer-vision model (software that "sees" images) identifies key body postures from video frames, hitting 99% accuracy. A second tool sorts the in-between movements, reaching 84%. Then an algorithm borrowed from spell-check compares the dancer's sequence with the ideal one and scores the match — 98% accurate. In plain terms: it's like autocorrect, but for dance steps instead of words.

Why does this matter outside a lab? Bharatanatyam is usually taught person-to-person, and learners far from a teacher get little feedback between lessons. A phone app built on this research could watch a practice session and flag a wrong step, or a step done out of order. The authors also frame it as "digital heritage" — using AI to record and preserve a centuries-old art form in a precise, searchable way, so it isn't lost as teachers age.

The catch: the system judges only shapes and order, not artistry, expression, or how well a dancer moves with the music — the parts audiences actually feel. It's also research code, tested on a limited set of recordings, and not yet a polished app. Think of it as a helpful practice partner, not a replacement for a guru.