A quadruped robot developed by the Robotic Systems Lab de ETH ZurichIn Switzerland, he has demonstrated his ability to compete in the dynamic sport of Badminton against human players.
As reported by the magazine Science Roboticsthe robot, called ANYmal-Duses a reinforcement -based control system that allows you to identify, follow and respond to the wheel trajectory (Shuttlecock) with remarkable precision, marking a significant advance in autonomous mobile robotics.
The design of ANYmal-D It includes a stereo camera and a dynamic arm equipped with a racket, in addition to a four -legged locomotion system that provides greater stability and flexibility compared to humans, who depend on two lower extremities.
During the tests performed, the robot managed to keep balls of up to ten consecutive blows with human players, adjusting their posture and displacement to intercept and return the steering wheel efficiently. This performance was possible thanks to a reinforcement learning controller that integrates locomotion, visual perception and real -time manipulation.
The researcher Yuntao Mamember of the development team, he explained that the system incorporates a noise perception model, which allows the robot to adjust its behavior based on the quality of visual perception. This approach automatically balances the agility of the robot movement with the reliability of its perception.
“We introduced a noise perception model that links the movement of the robot with the quality of perception. This allowed the learning algorithm to automatically balance between the agile movement of the robot and a reliable perception,” MA told Science Robotics.
In terms of performance, ANYmal-D He reached monitoring speeds of up to 12.06 meters per second and showed the ability to rise on his hind legs to improve the visibility of the steering wheel.
In addition, he adjusted his balance dynamically to optimize his performance in the game. However, the system presented limitations against aggressive blows, such as the “smokes”, due to the speed of its actuators and a response latency of 0.375 seconds.
The researchers acknowledged that it will be necessary to improve the perception speed so that the robot can compete effectively in full court matches.
The study, published in Science RoboticsHe stressed that the control algorithm was trained unified, without separating the functions of the upper and lower extremities.
This allowed the robot extremities to learn to compensate for each other, achieving a complete body movement coordination. According to ma, the principles of this development could be applied to other robotic platforms, such as humanoids, and extending to tasks beyond sport, including rescue operations and domestic services.
“The control algorithm is also generalized to other robotic platforms, such as humanoids, and tasks such as search and rescue or services at home,” said the researcher.
The success of ANYmal-D In a fast and variable sports environment such as the Bádminton underlines the potential of robots to operate in dynamic and complex tasks. The integration of locomotion, visual perception and real -time manipulation through automatic learning reinforces the viability of using robots with multiple limbs in scenarios that require high adaptability, both in recreational and functional contexts.
Worth a look
