Robot vs human – badminton gets a bionic opponent

A four-legged robot developed by ETH Zurich can play real-time badminton against humans, using reinforcement learning and vision-based motion control to track, move, and return shots with striking accuracy.

The sweat is dripping down your forehead as you move around the badminton court. You’ve been working on your “overhead clear” shot for a while now.

You remember the 90-degree angle to position your feet, follow your leading arm through, and connect the racket with the shuttlecock.

Not many people would be able to return such a shot, but on the other side of the net isn’t a human – instead, it’s a four-legged robot that’s quite capable of smashing it back.

This isn’t sci-fi. It's the result of cutting-edge robotics research from ETH Zurich, which has designed a bot that can play badminton with stunning accuracy.

When robots rally back

The researchers developed a robot, named ANY-mal-D, capable of playing in real-time against humans.

It autonomously tracks a shuttlecock, navigates a court, and uses a dynamic arm to return shots.

During testing, the robot maintained rallies of up to 10 strokes, responding to different speeds and angles.

Its ability to adapt mid-rally shows real-time decision-making and physical agility rarely seen in legged robots.

Perception in motion

“Our system allows the robot to make rapid, full-body decisions in real time,” said lead author Yuntao Ma, “which is essential for fast-paced, interactive tasks like badminton.”

The robot is powered by a reinforcement learning-based controller trained to link perception with movement.

The engineers have equipped the badminton-bot with a stereo camera to detect a shuttlecock’s trajectory.

In particular, the machine coordinates leg motion, body posture, and arm swings to respond rapidly and accurately.

Beyond the court

Usually, achieving real-time athletic movement is notoriously difficult due to coordination challenges.

In warehouses, for example, despite being able to work longer hours, robots don’t always have the same agility as humans.

The “whole-body approach” that is being used here needn’t be confined to the badminton court.

The bots could be utilized for disaster response, elder care, or dynamic warehouse tasks.

Yuntao Ma and team call this a “template” for deploying athletic, perception-guided robots in dynamic environments.

And that could be the winning shot, as robots fully adapt to share human spaces.