Robot Drummer executing a cymbal (dark brown) strike in a learned rhythmic sequence.

Humanoid robots, robots with a human-like body structure, have so far been primarily tested on manual tasks that entail supporting humans in their daily activities, such as carrying objects, collecting samples in hazardous environments, supporting older adults or acting as physical therapy assistants. In contrast, their potential for completing expressive physical tasks rooted in creative disciplines, such as playing an instrument or participating in performance arts, remains largely unexplored.

Researchers at SUPSI, IDSIA and Politecnico di Milano recently introduced Robot Drummer, a new humanoid robot that can play the drums both accurately and expressively, supported by a reinforcement learning algorithm. This robot, presented in a paper published on the arXiv preprint server, was found to gradually acquire human-like behaviors, including movements that are often performed by drummers.

"The idea for Robot Drummer actually emerged from a spontaneous conversation over coffee with my co-author, Loris Roveda," Asad Ali Shahid, first author of the paper, told Tech Xplore. "We were discussing how humanoid robots have become increasingly capable at a wide range of tasks, but rarely engage in creative and expressive domains. That raised a fascinating question: what if a humanoid robot could take on a creative role, like performing music? Drumming seemed like a perfect frontier, as it's rhythmic, physical, and requires rapid coordination across limbs."

The main objective of this recent study by Shahid and his colleagues was to develop a machine learning-based system that would allow a humanoid robot to play the drums for entire musical tracks, exhibiting rhythmic skills similar to those of human performers. The system they developed, dubbed Robot Drummer, was successfully evaluated in simulation on the G1 humanoid robot, created by Unitree, allowing it to accurately play real songs on a drum kit.

"The key idea is that we represent each piece of music as a chain of precisely timed contact events, what we call a rhythmic contact chain," explained Shahid.

"These contact goals tell the robot which drums to strike and when. The robot then uses this information and learns how to hit the drums by practicing and improving in a simulated environment. What's unique is that the robot develops intelligent behaviors, like dynamically switching sticks, crossing its arms to reach different drums, and optimizing its movements across the kit to adapt to the rhythm—in ways that resemble human drummers."

To evaluate their system, the researchers tested it on a simulated G1 Unitree robot, assessing its ability to perform popular songs across different genres, from jazz to rock and metal. These included well-known tracks like "In the End" by Linkin Park, "Take Five" by Dave Brubeck and "Living on a Prayer" by Bon Jovi. They found that the robot effectively learned complex rhythmic structures and played the songs with high accuracy, often with a rhythmic precision of over 90%.

The humanoid robot prepares to strike a snare drum (green).

"More impressively, the robot discovered human-like strategies, such as planning for upcoming strikes, executing cross-arm hits, and reassigning sticks on the fly," said Shahid. "These behaviors emerged entirely by optimizing for rhythmic rewards the robot received during training. In the long term, Robot Drummer could pave the way for robotic performers to accompany live bands on stage, and it provides a framework for teaching precise timing skills in domains beyond music."

This recent study could soon inspire other research groups to develop machine learning-based tools that allow humanoid robots to play musical instruments or engage in other performance arts. Robot Drummer and other similar systems could prove valuable for the entertainment industry and could eventually be deployed in real-world settings to showcase advancements in robotics.

"Our next step is to bring Robot Drummer into the real world by transferring learned skills to actual hardware," added Shahid. "We also aim to teach the robot to improvise and adapt its playing style instead of following a 'fixed drum sheet.' For example, we could enable the robot to adjust its timing on-the-fly based on musical cues— essentially giving it the ability to 'feel' and respond to music like a human drummer."

More information: Asad Ali Shahid et al, Robot Drummer: Learning Rhythmic Skills for Humanoid Drumming, arXiv (2025). DOI: 10.48550/arxiv.2507.11498

Journal information: arXiv