Researchers at the ETH Zurich State Research University are developing a four-legged ANYmal robot and teaching it to play badminton.

The developers have set themselves a challenging task - coordinating the movements of both upper and lower limbs. To achieve this, they use a mobile manipulation system that engages all of the robot’s joints, along with its built-in sensors.

As a result of this training, the quadruped robot can move freely across the court, strike the shuttlecock at various angles, and participate in extended matches with human players.

Thanks to its integrated sensors, the robot can predict the shuttlecock’s trajectory by filtering data based on color and applying an aerodynamic model. The reward system minimizes the likelihood of errors.

The quadruped robot also features an adaptive walking function that adjusts to different conditions. Depending on the task, one leg can remain in a supporting role for an extended period, the robot can move at a gallop, and it can switch strategies under time constraints by adapting to more suitable movement modes.

As practice shows, the potential of legged manipulators is significant. The agility and precision required for playing badminton can also be applied in other fields where minimizing risk is critically important.