Ballerina turns to AI

For Milka Trajkova, technology and dance are part of the same rhythm. A ballerina turned researcher, she’s using data science to reveal how movement can be measured and refined.

Her work analyses motion data to give dancers and teachers clearer insight into technique, helping prevent injury and improve precision.

What began as a five-year-old’s passion for ballet has evolved into research that blends human-computer interaction with the art of performance.

Act I: Ballet beginnings

Trajkova first laced up her ballet shoes at age 5, when her family still lived in Macedonia. Her instructors realised she had an aptitude for dance, but her family immigrated to America the following year, and she lost interest.

One year later, Trajkova wanted to get back on the dance floor. The costumes, classical music, and storytelling thrilled her. And while she chasséd through the DC suburbs growing up, Trajkova regularly went back to Macedonia and kept in touch with the ballet company there.

When she was 17, Trajkova was given the opportunity to join her childhood company, the Macedonian Opera and Ballet, as part of the corps de ballet. She moved back to Macedonia to dance.

Despite achieving her childhood dream, Trajkova knew she needed a backup plan. So, she enrolled in the only accredited English-language program at the capital’s Southeast European University (being more fluent in English than her native tongue).

The programme was in information systems management (ISM), offered in collaboration with the London School of Economics. In the mornings, she took classes, then rushed to the theatre for warm-up class and afternoon rehearsals, and finished her studies in the evening.

“Both of my parents are engineers, and I always said growing up, ‘Whatever you do with computers, I never want to do that,’” says Trajkova. “But that was the only programme available, so I just dove into it.

“HCI is a fusion of design, psychology, and technology, and it opened my eyes. HCI is about understanding people first, then shaping technology around them – so technology doesn’t just work, it works for us.

“It was like opening a door to another world. That’s when things really took off, and I found what I wanted to study.”

Trajkova found a paper about a computer program that could assess whether dancers were performing ballet movements correctly. She reached out to the original researcher and asked if she could user-test the program, having access to research subjects at the ballet school where she practiced daily.

Act II: From dance to data

Trajkova, aged 17, gave her first performance of Swan Lake for the Macedonian Opera and Ballet

As Trajkova was discovering HCI and working on her research, she developed a debilitating toe injury. Doctors warned her she would have to endure constant pain if she wanted to keep performing. At the age of 20, after only three years at the company, Trajkova retired from ballet.

“It was very hard to deal with for a long time. But my research gave me a purpose again, and I wanted to give back to an art form that gave me everything.”

Her undergraduate thesis on the plié was her opening act. Trajkova then pursued both a master’s degree in HCI and a PhD in informatics at Indiana University. She wanted to quantify ballet movements as data to train an AI model. From there, she believed the AI program could be used to enhance dancers’ technique and performance.

“When you walk into a ballet class, the highest form of technology dancers use daily is a mirror,” she says. “I wanted to change that.”

Ballet is often perceived as more of an art form than an athletic pursuit worth quantifying. Yet behind every seemingly fluid movement is correct technique. Breaking down technique from a motor-learning perspective can pinpoint where dancers go wrong. Using this data, Trajkova created aiDance, a dashboard that helps dancers build visual awareness of their technique and gain actionable insights to improve performance.

The idea of using AI at the ballet barre initially made her fellow dancers skeptical, but Trajkova’s background helped. She wasn’t a researcher who merely had seen one production of the Nutcracker; she had been in those same satin pointe shoes.

“People like to create a kind of mystery around the ballet world, when, at the end of the day, dancers are athletes who need to be well-equipped to perform any role. They need to know how they can refine their technique and maintain their body for the growing demands of the art form.”

Preventing injury is the emphasis of Trajkova’s research, but she also wants to aid ballet teachers as much as students. In a typical ballet class, dozens of dancers share the attention of just one teacher, making personalised and individualised feedback nearly impossible. Trajkova says her work is a way to supplement, not replace, teachers. It aims to help them gain a deeper understanding of each student’s progress and needs.

She conducted most of her PhD research during the Covid-19 pandemic, when her ballet conservatory in Maryland recorded all classes on Zoom and Microsoft Teams. From those videos, she was able to extract meaningful features of ballet motion and train a machine learning model to identify a specific error in a plié.

Act III: Choreographing a future

During her PhD program, Trajkova met IAC Professor Brian Magerko at an academic conference. Magerko was also doing dance and AI work, and Trajkova knew he had to be on her dissertation committee. Similarly, Magerko thought Trajkova would be a great fit at Georgia Tech. In 2021, she interviewed for her current research scientist role and arrived on campus shortly after.

At Tech, Trajkova changed her dance research from ballet to improvisational movement. Magerko’s lab was designing an improvisational dance partner called LuminAI, which uses AI to transform human movement into new movement.

This fosters a co-creative exchange between human and machine. The project, funded by the National Science Foundation, was the first to culminate in both a long-term improvisational dance class study and a public performance of AI-driven improvisation.

Trajkova’s research has further widened in scope; now she wants to decode a broad spectrum of artistic movement. From rhythmic gymnastics to ballet, adding data to “artistry” has been nearly impossible until AI.

In 2024, Trajkova hosted a workshop with 12 experts from around the world representing various disciplines, including human-computer interaction, data visualisation, bioinformatics, music and robotics, ballet, biomechanics, sports science, and gymnastics. The group met to discuss how they could begin this analysis.

Dance as an art form isn’t centre stage in this work. There are other applications, including healthcare. Dance therapy can be used to help with Parkinson’s disease rehabilitation, for example.

“Most physical rehabilitation has been focused on function, like picking up a cup,” says Trajkova. “We’re asking how we can reframe rehabilitation to make it feel more artistic and enjoyable for people – because the more engaging it is, the more it can support neuroplasticity and lasting recovery.”

“My life has come full circle because I started with ballet, and now I can contribute to dance in my own way. It’s rewarding that I still get to be part of the dance community and interact with dancers.”