Meet the Drumming Robot: Scientists are teaching AI to play drums in the style of Linkin Park and AC/DC, but it looks like it still needs a lot of practice.

In the End – Linkin Park – Robot Drum Cover – YouTube

Anthropomorphic machines can manipulate objects, move quickly, and even engage in physical activity (within certain limits), but can they master the rhythm of percussion instruments? New research suggests that it is, in principle, possible.

The virtual drummer is a humanoid avatar that delivers “emotional, technically precise performance,” according to researchers from the Polytechnic University of Milan in a study published July 15 on the preprint archive ArXiv.

The idea for the project, according to co-authors Asaf Ali Shahid and Loris Roveda in Tech Explore, was born by chance during a coffee break.

You might be interested

China to Host First Global 'Roboboxing Championship' Between Anthropomorphic Machines
Unique 3D Printed Elephant Robot Shows Off Bowling Skills In New Video
Aviation pioneer: humanoid with jet engines takes to the air for the first time

“We were thinking about the increased capabilities of anthropomorphic systems, which are rarely used in creative and artistic fields,” Ali Shahid said. “This sparked a discussion: what if a machine could perform creative tasks, such as playing music? The drum kit was an ideal choice because of the need for synchronization of movements, rhythm, and physicality.”

The virtual drummer is not a physical device, but a model based on the G1 Unitree robot, which could potentially be used by humanoids. The 3D animation shows the G1 interacting with the drum kit's color indicators.

The YouTube video platform features a demo of a robot playing rhythmic patterns of two well-known compositions. The sound contains some peculiarities, perhaps to make it easier to understand. Linkin Park's version of the track exhibits minor rhythmic deviations, while The Police's interpretation of “Roxanne” sounds more coherent.

According to researchers, mastering a musical instrument, especially percussion instruments, is associated with complex difficulties. It requires quick reactions, precise interaction and coordinated work of the limbs throughout the performance.

The system's training was based on a trial-and-error method using reinforcement learning (RL), an AI technology in which an algorithm improves strategies through interaction with the environment.

CURRENT MATERIALS

— MIT's artificial intelligence is learning to control cars through visual perception — all it takes is a single camera.

— Scientists are testing innovative “sensitive” robotic skin that is resistant to damage.

— Meta-AI Takes a Step Towards Superintelligence — Zuckerberg to Limit Public Access to Most Advanced Systems

The algorithm receives numerical scores for actions that stimulate the search for optimal solutions. For the system, each composition was transformed into a chain of precisely synchronized “pulses” indicating the time and location of the strikes.

The scientists identified techniques they mastered: cross-strokes and adaptive stick distribution to create specific sounds. This confirms the potential of the RL approach for creative applications of robotics.

The system has been tested on 30+ popular tracks in the styles of pop, rock, metal and jazz.

Bobby Hellard

To participate in the discussion, please confirm your public profile name.

Please log out and log back in to enter your display name.

Logout Read more