Anymal-D Robot Competes in Badminton
Engineers at ETH Zurich’s Robot Systems Lab have developed Anymal-D, a remarkable four-legged robot capable of engaging in badminton matches with human players. This integration of robotics, artificial intelligence, and sports showcases how sophisticated technology can participate in dynamic and fast-paced games.
The design and features of Anymal-D open doors for new forms of collaboration between humans and robots in sports and other areas.
How Anymal-D Plays Badminton
Badminton demands quick reflexes, accurate footwork, and precise coordination. To give Anymal-D a competitive edge, the ETH Zurich team equipped the robot with four legs for stability, a dynamic arm for swinging a racket, and a stereo camera for tracking the shuttlecock. Its reinforced learning-based controller allows the robot to predict and react to the shuttlecock’s movements in real time, adapting its position as it plays up to ten shots with human players.
The Technology Behind Anymal-D’s Skills
Anymal-D constantly watches the shuttlecock through its stereo camera. It employs a “Perceptual Noise Model” to match its training data with real-world scenarios, enabling it to track the shuttlecock even during unpredictable movements. The robot can shift its body and mimic the angles humans use for tricky shots.
Unified Control for Smooth Movements
Coordinating the movements of legs and arms is particularly challenging for robots. The ETH Zurich team developed a unified control policy through reinforcement learning, allowing Anymal-D to operate as a cohesive unit. The robot underwent extensive training in simulations, preparing it for various shooting situations before entering real matches.
Analyzing the Robot’s Hardware
Anymal-D features a robust rectangular base with a Dynaarm, positioning its racket at a 45-degree angle for optimal striking. Its various systems operate at differing frequencies, ensuring high responsiveness and readiness for play.
Challenges of Playing Badminton with a Robot
Linking the robot’s arms and legs smoothly presents a significant challenge, as most robots traditionally handle these independently, limiting agility. However, by combining both movement and arm control, Anymal-D adapts its posture and walking to the shuttlecock’s trajectory, closely resembling human play.
Active Recognition: How Anymal-D Interacts with the Game
As robots lack human-like vision, issues like frame rates and field of view can affect performance. Anymal-D’s recognition system ensures its cameras maintain smooth tracking of the shuttlecock, utilizing perception noise models to enhance reliability during matches.
Real-World Development of Anymal-D
Transitioning Anymal-D from the lab to actual badminton courts involved overcoming obstacles such as power restrictions and communication delays. Nevertheless, the robot showcased its ability to match human players in terms of speed and adaptability, effectively engaging with various shot speeds and landing margins during rallies.
Performance Insights from Anymal-D
In collaborative games with amateur players, Anymal-D displayed impressive consistency in tracking, intercepting, and returning shuttlecocks. It averaged 0.357 seconds to process trajectories after hits, demonstrating a commendable ability to keep the rally going, illustrating the advancements in robotics for dynamic sports contexts.
Conclusion
Anymal-D exemplifies the remarkable progress in robotics, particularly in high-speed activities like badminton. Watching this robot adapt and interact with human players is fascinating and hints at a future where robots could engage in diverse sports, enhancing enjoyment and teamwork for everyone involved.
