homehome Home chatchat Notifications


Meet the Robot Drummer That Can Play Linkin Park (and Bon Jovi) Like a Human

Robots can play music while we work our menial jobs.

Mihai Andrei
August 12, 2025 @ 7:53 am

share Share

Robot Drummer executing a cymbal (dark brown) strike in a learned rhythmic sequence. Credit: Shahid, Braghin & Roveda.

We associate robots with useful, practical activities. They carry things, help us, or provide some form of utility. But Robot Drummer is different.

The robot, developed by scientists in Switzerland and Italy can drum full songs — rock, jazz, even metal, all with striking accuracy. It’s not just a beat machine, though. It learns human-like tricks, from cross-arm hits to on-the-fly stick swaps. It’s a fresh step toward robots that can perform creative, physical arts, not just lift boxes.

Robot Goes Brr

The team used a Unitree G1 humanoid in simulation, perhaps the most performant commercially available humanoid robot. A reinforcement-learning algorithm rewarded the robot for hitting the right drum at the right time, and penalized misses or wrong hits. A gentle bonus encouraged moving sticks close to upcoming targets at the right moments. This dense reward design teaches timing and coordination — not just “did you hit it,” but “did you hit it on time.”

Instead of learning from audio, the robot read the drum part of a song as a sequence of precisely timed “contact events.” The researchers call this a Rhythmic Contact Chain. Each event says which drum to hit and when, and the robot then learns how to move its two sticks and torso to hit those targets in time. Songs are split into short segments, and one learning policy trains across all segments in parallel. That keeps learning efficient without losing the long-range flow of music.

In tests across 30+ songs, Robot Drummer (how did they come up with the name?) nailed the rhythm. On tracks like “In the End” (Linkin Park), “Livin’ on a Prayer” (Bon Jovi), “Take Five” (Dave Brubeck), “Fire” (Jimi Hendrix), and more, it only got a few misses and imperfections per song. It also handled fast snare rolls and complex fills, and discovered efficient moves such as cross-overs and stick reassignment without being told to do so.

The robot struggled with irregular timing. Songs with uneven spacing between hits were harder, and frequent rhythm changes also hurt its performance. Tempo alone didn’t seem to matter much; faster or slow, the robot got it right as long as there were no variations.

Perhaps most impressively, the robot discovered human strategies by itself. It figured out cross-arm hits are more efficient sometimes, and executed seemingly counterintuitive moves much like humans.

People Think It Sounds Pretty Creative

To evaluate its expressiveness, researchers conducted a small-scale listener study. They had 15 listeners watch and listen to at least three songs performed by the robot and rate various aspects of the performance on a scale from 1 to 5. Unsurprisingly, the “Naturalness” scored the lowest, just under 2.5, while its expressiveness fared pretty well (3.4). For comparison, Timing and Consistency scored 3.7. The performance is good, but the vibe is still “robot”.

Most humanoids focus on walking, balancing, and simple manipulation. Music is different. It demands split-second timing, fast multi-limb coordination, and endurance over minutes. Even tiny timing errors ruin the groove. The team frames drumming as a process, not a single goal — something robot learning has struggled with.

Robots have played instruments before, usually with fixed motions or simplified arms. Reinforcement learning has taught dexterous hands to play piano pieces, but coordinating a full humanoid body over a drum kit, with timing, reach, and two sticks, is a leap. This study shows that framing music as a chain of timed contacts plus dense, musically meaningful rewards can unlock expressive, long-horizon skills. That approach could transfer to dance, sports drills, or any task where what you do next depends tightly on when you do it.

The system runs in simulation today. The team plans to move to real hardware, likely starting with an electronic kit that reports MIDI hit timing and strength. That feedback could fine-tune the policy in the real world and close the sim-to-real gap. They also want to train the robot to improvise and adapt its timing and feeling rather than following a fixed cue.

So, would you go to a robot concert?

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

share Share

Birds Are Singing Nearly An Hour Longer Every Day Because Of City Lights

Light pollution is making birds sing nearly an hour longer each day

U.S. Mine Waste Contains Enough Critical Minerals and Rare Earths to Easily End Imports. But Tapping into These Resources Is Anything but Easy

The rocks we discard hold the clean energy minerals we need most.

Scientists Master the Process For Better Chocolate and It’s Not in the Beans

Researchers finally control the fermentation process that can make or break chocolate.

Most Countries in the World Were Ready for a Historic Plastic Agreement. Oil Giants Killed It

Diplomats from 184 nations packed their bags with no deal and no clear path forward.

Are you really allergic to penicillin? A pharmacist explains why there’s a good chance you’re not − and how you can find out for sure

We could have some good news.

Archaeologists Find 2,000-Year-Old Roman ‘Drug Stash’ Hidden Inside a Bone

Archaeologists have finally proven that Romans used black henbane. But how did they use it?

Astronomers Capture the 'Eye of Sauron' Billions of Light Years Away and It Might Be the Most Powerful Particle Accelerator Ever Found

A distant galaxy’s jet could be the universe’s most extreme particle accelerator.

Scientists Have a Plan to Launch a Chip-Sized, Laser-Powered Spacecraft Toward a Nearby Black Hole and Wait 100 Years for It to Send a Signal Home

One scientist thinks we can see what's really in a black hole.

What Would Happen If Everyone in the World Turned On The Lights At the Same Time?

Power grids could likely handle the surge of demand, but all that light would pollute dark zones nearby.

AI Designs Computer Chips We Can't Understand — But They Work Really Well

Can we trust systems we don’t fully understand?