ANYmal, an advanced robotic system, has successfully learned to avoid falls and assess acceptable levels of risk based on its limited speed. While it displayed a commitment to the game, it refrained from executing impossible shots that could lead to potential damage.
However, when pitted against human rivals, ANYmal’s performance as a badminton player was subpar, demonstrating amateur-level skills.
The Major Leagues
The primary challenge for ANYmal was its slow reaction time. Average humans respond to visual stimuli in approximately 0.2 to 0.25 seconds. Elite badminton players, equipped with sharp reflexes, anticipation, and muscle memory, can lower this response time to between 0.12 and 0.15 seconds. In contrast, ANYmal required about 0.35 seconds after an opponent struck the shuttlecock to recognize trajectories and decide on a response.
Another contributing factor was the robot’s limited visual capabilities. Researcher Ma noted, “Perception remains a significant hurdle.” ANYmal utilized a stereo camera to locate the shuttlecock, but this method came with potential positional errors at each time interval. The camera’s narrow field of view further limited the robot’s ability to track the shuttlecock for extended periods. Ma remarked that “it was better suited for less intense matches—when the human player begins to smash, the robot’s success rate significantly decreases.”
Nevertheless, the team has outlined strategies to enhance ANYmal’s performance. They aim to improve reaction times by enabling the robot to predict the shuttlecock’s trajectory based on the opposing player’s body language, a tactic often employed by seasoned badminton and tennis players. Additionally, they plan to equip ANYmal with advanced technology like event cameras, which can capture motion with ultra-low latency. Enhancements may also include faster and more efficient actuators.
Ma expressed optimism about the broader applications of their training framework, suggesting it could be helpful in various scenarios where perception and control need to be balanced, such as picking up objects or even catching and throwing items. However, he conceded that the idea of introducing ANYmal into professional badminton or tennis leagues is unlikely. “Would I establish a company focused on selling badminton-playing robots? Probably not,” he concluded.
Science Robotics, 2025. DOI: 10.1126/scirobotics.adu3922
