Robots are learning new skills far beyond factories and labs. In a badminton hall, one four-legged machine is now chasing shuttlecocks and rallying with human opponents, thanks to artificial intelligence.
How did the robot learn to play badminton
Researchers trained a quadruped robot named ANYmal to play rallies of up to ten shots. The dog-like machine weighs around 50 kilograms and usually stands half a metre tall. With a robotic arm and racket attached, it reached 1.6 metres in height. By using visual perception and whole-body movement, it learnt to track the shuttlecock and return it successfully.
The team built a simulation of a badminton court and ran 50 million trial sessions. A neural network was established to manage 18 joints, four legs and the racket arm. The robot was rewarded for racket position, swing angle, speed, and movement efficiency. These strict rules helped it develop a playing style similar to humans.
What happened during real-world testing
After training, the neural network was transferred to the actual robot. ANYmal faced shuttlecocks served by a machine at different speeds and angles. The robot learnt to scuttle, scramble, and even gallop to return shots. Swing speeds reached 12 metres per second, about half that of an amateur human player.
The researchers equipped ANYmal with a four-legged frame, a long robotic arm holding a badminton racket at a 45-degree angle, and a stereo camera mounted at the front. (Image: © 2025 Yuntao Ma/ Robotic Systems Lab/ ETH Zurich.)
ANYmal adapted its movement based on distance and timing. At short distances, it stayed still, while longer shots forced it to sprint across the court. The robot even developed the ability to go back to the centre after every hit, just like human tactics.
What challenges and future uses remain
The robot is still unable to anticipate an opponent's movement, which restricts its capacity to foresee shuttlecock trajectories. Adding pose recognition and a flexible neck joint could improve performance. The research team says these skills may one day support disaster relief operations, where robots must balance vision and agility to remove debris safely.
“This is not just about sport,” said researcher Yuntao Ma, who led the work. “It shows how a robot can coordinate complex movements with vision in real time.”
Discover the latest Business News, Sensex, and Nifty updates. Obtain Personal Finance insights, tax queries, and expert opinions on Moneycontrol or download the Moneycontrol App to stay updated!
