This Robot Now Plays Badminton With You — Jump Smashes Included

A team of researchers has taught a humanoid robot — a machine built to look and move like a person — to play badminton against real humans. That sounds like a party trick, but badminton is brutally hard for robots. The shuttlecock moves fast, curves unpredictably, and must be hit with the right racket angle and speed within a window of a few milliseconds, often while the robot is running or jumping.

The trick was how they trained it. Instead of hand-programming every swing, they used reinforcement learning (AI that learns by trial and error, like a kid practising). Because examples of human badminton motion are scarce, they stretched a small set of human swings into thousands of practice variations. A 'planner' AI then picks the right shot in real time based on where the shuttle is heading, while a second AI keeps the robot's movement looking natural instead of awkward.

On a real robot, the system held long rallies with human players. It hit forehands, backhands and — the standout — leaping jump returns, where the robot launches into the air to strike the shuttle. The authors say this is the first real-world humanoid racket-sport system to manage multi-skill rallies with people, including those jumping shots. The work was accepted at CoRL 2026, a leading robotics conference.

The catch: this is one robot, one sport, likely on a controlled court, and no price tag was mentioned — humanoid robots today cost as much as a car or a house. The bigger story is the underlying skill set: split-second decisions, precise physical timing and whole-body balance. Those are the same abilities needed for warehouse work, helping in care homes, or being a practice partner who never gets tired.