Zurich, Switzerland - Researchers at ETH Zurich and Science Robotics have developed a badminton-playing robot that uses reinforcement learning, a type of AI that improves decision-making through repeated trials.
The robot learns through trial and error to make better decisions. The AI technology combines vision, movement, and hand control, enabling the robot to track and accurately hit a "shuttlecock."
A noise detection model trained with real-world camera data ensures the robot's consistent performance. Yuntao Ma, a member of the research team, stated, "We presented a noise detection model that equates the robot's movement with the quality of perception. This allowed the reinforcement learning algorithm to automatically balance the robot's agile movement with reliable perception."