How do you get a robot to move, see, and hit a shuttlecock back—all at the same time? A team of researchers at ETH Zurich, led by Professor Marco Hutter, has been working on that exact challenge.
Their quadruped robot, ANYmal, is best known for tasks like climbing stairs, navigating rubble, and performing inspections in industrial environments. But in a new twist, the team has trained it to step onto the badminton court.
The secret lies in a new control system that synchronizes the robot’s leg movements, racket strokes, and camera vision. ANYmal uses its two cameras to track the shuttlecock in real time, predicts its flight path, and quickly navigates to the right position to return the shot.
In the accompanying video, Andrei Cramariuc, a postdoctoral researcher on Hutter’s team, breaks down the training process. He highlights the challenges of teaching the robot to anticipate fast, unpredictable shuttle trajectories while still maintaining balance and timing its racket swing.
Beyond the novelty of a robot dog playing badminton, the project demonstrates the complex coordination between perception, prediction, and action required for high-speed, dynamic tasks. These are the same skills robots will need for real-world applications like disaster response, agile manufacturing, and human-robot collaboration.
