Science Robotics, 2025
Yuntao Ma [1], Andrei Cramariuc [1], Farbod Farshidian [2], Marco Hutter [1]
[1] Robotic Systems Lab, ETH Zürich [2] Currently at Robotics and AI Institute
Coordinating the motion between lower and upper limbs and aligning limb control with perception are substantial challenges in robotics, particularly in dynamic environments. To this end, we introduce an approach for enabling legged mobile manipulators to play badminton, a task that requires precise coordination of perception, locomotion, and arm swinging. We propose a unified reinforcement learning-based control policy for whole-body visuomotor skills involving all degrees of freedom to achieve effective shuttlecock tracking and striking. This policy is informed by a perception noise model that utilizes real-world camera data, allowing for consistent perception error levels between simulation and deployment and encouraging learned active perception behaviors. Our method includes a shuttlecock prediction model, constrained reinforcement learning for robust motion control, and integrated system identification techniques to enhance deployment readiness. Extensive experimental results in a variety of environments validate the robot's capability to predict shuttlecock trajectories, navigate the service area effectively, and execute precise strikes against human players, demonstrating the feasibility of using legged mobile manipulators in complex and dynamic sports scenarios.
Selected Supplementary Videos
To evaluate how precisely the robot can follow swing velocity commands, we conducted controlled swing tests. This video shows the test setup and some high-speed racket swings.
Our method is not limited to quadrupedal morphologies—it generalizes to humanoid robots as well. This video demonstrates a control policy trained on the Unitree G1 humanoid model. Except for adjusting the head camera orientation in simulation, all robot specifications, including actuator limits, are respected.
To build the perception noise model, we recorded the shuttlecock’s position using the robot’s stereo camera in a motion capture environment. This allowed us to quantify how factors like distance and camera shake impact perception accuracy. The resulting model was then used during training to simulate realistic perception noise and balance the trade-off between agility and perception reliability.
Citation
@article{ma2025badminton,
author = {Yuntao Ma and Andrei Cramariuc and Farbod Farshidian and Marco Hutter},
title = {Learning Coordinated Badminton Skills for Legged Manipulators},
journal = {Science Robotics},
volume = {10},
number = {102},
pages = {eadu3922},
year = {2025},
doi = {10.1126/scirobotics.adu3922},
url = {https://www.science.org/doi/10.1126/scirobotics.adu3922}
}