Legged robots that play badminton and do parkour can coordinate perception and movement.

Two scientific teams, one in Switzerland and the other in South Korea, have developed control frameworks that have enabled legged robots to execute parkour-like movements and participate in badminton matches against humans . These frameworks—a structured set of algorithms, rules, and methods that guide the robots’ movement and behavior—combine planning, reinforcement learning, and real-time perception strategies to maneuver autonomously in complex or dynamic terrain.

In one of the experiments—reported in the journal Science Robotics —a quadruped robot named Raibo demonstrated its agility by running up vertical walls, leaping over large gaps, and climbing stairs at high speed. The framework implemented in this machine uses a neural network to optimize safe footings and a map generator to simulate terrain of varying difficulty, allowing the robot to adapt to diverse physical environments.

On the other hand, another four-legged robot, the ANYmal-D , used this type of control to play badminton against humans. Equipped with a stereo camera and a dynamic arm, the robot learned to track and predict the trajectory of the shuttlecock, moving around the court to intercept and successfully return it. Tests showed it could sustain rallies of up to 10 consecutive shots, always prioritizing the stability of its posture.

These advances could lay the groundwork for designing robots that can navigate autonomously in disaster or construction sites, and for manipulating objects in dynamic tasks that require coordination, rapid perception, and agile movements.