Robots playing badminton? It’s not just a party trick, it’s a breakthrough in whole-body visuomotor control. Researchers from ETH Zürich have trained a legged mobile manipulator to rally with human players using a unified reinforcement learning policy that tightly coordinates perception, locomotion, and arm movement. Key innovations: – Real-world-informed perception noise model – Constrained RL for robust motion – Active perception + shuttlecock prediction The result? A robot that can track, navigate, and smash, all while adapting in real time to a dynamic game environment. 📍 Another step toward truly agile, perceptive, and adaptable robots.