A robot that can freely play badminton with people has emerged. It must be able to control its entire body freely and perform various perceptual functions integratively, which indicates a high potential for application in disaster and emergency situations as well as sports robots.
A research team led by Professor Marco Hutter from the Department of Mechanical Engineering at the Swiss Federal Institute of Technology Zurich (ETHZ) published the results of their study on a quadruped robot capable of autonomously playing badminton in the international journal 'Science Robotics' on the 29th.
The research team modified the quadruped robot 'ANYmal-D' to play badminton. They equipped it with stereo cameras to recognize incoming shuttlecocks and attached robotic arms to its upper body to allow it to swing a badminton racket.
The robot succeeded in maintaining a rally of up to 10 times while playing badminton with a person. Even when the person sent shuttlecocks at different speeds and angles, the robot moved around the court and retrieved the shuttlecock. The robot even demonstrated the ability to stand on its hind legs to avoid losing sight of the shuttlecock. Its swing speed was about 12.06 meters per second, which is on par with amateur players.
The secret to the robot's ability to play badminton with humans lay not in its limbs but in its head. Instead of programming it to follow commands, the research team allowed the artificial intelligence (AI) to learn badminton on its own. Researcher Yuntao Ma, who led the study, explained, "The technology allows the robot to track and predict the trajectory of the shuttlecock and accurately return it while moving around the court using a reinforcement learning-based control strategy."
AI reinforcement learning is a training method that rewards a dog when it accidentally performs a specific action, rather than continuously teaching it to perform that action. This method is widely used when correcting a pet's behavior in television entertainment shows. When the robot accidentally hit the shuttlecock correctly, it was rewarded, enabling the AI to learn badminton on its own.
The research team noted that the robotic technology developed this time could be expanded into other sports fields, not just badminton. Controlling competitive robots is one of the most challenging areas in robotics due to the necessity of integrating cognitive functions with fast movement speeds and precise coordination of reactive actions. Researcher Yuntao Ma stated, "Developing a framework that integrates cognitive functions with the robot's upper and lower limb movements is a significant achievement."
In particular, the research team focused on the potential for advancement in badminton robots. To play badminton, the robot must move around the court and accurately receive falling shuttlecocks with its robotic arm. Researcher Yuntao Ma said, "The framework developed this time can also be applied to dynamic task control for humanoid robots or other multi-legged robots," indicating that the same technology could be utilized in rescue operations, emergency situations, or robots working alongside humans in factories.
The ANYmal robot that played badminton with humans was commercialized by ANYbotics, a company founded in Professor Hutter's laboratory. This robot won first place in the underground exploration robot competition hosted by the Defense Advanced Research Projects Agency (DARPA) in 2021 and also carried out remote inspection tasks in the sewers of Zurich and at a marine substation in the North Sea.
References
Science Robotics (2025), DOI : https://doi.org/10.5281/zenodo.15242151