A robot with legs can play the Badminton autonomously with humans, an indigenous study that describes the control strategy based on reinforcement learning – rama of artificial intelligence – that makes this feat possible.
The control and perception system (“brain” of the robot) allowed this to follow and predict the path of the steering wheel and move along the track to intercept and return it successfully.
Beyond the Bádminton, the method offers a template to display apparatus with legs in other dynamic tasks in which both the precise detection and the rapid responses of the entire body are fundamental, says Yuntao Ma, of the Federal Polytechnic School (ETH) of Zurich.
This scientist and his team publish the details in the journal Science Robotics.
Athletic robots control is a challenge, since it requires coordination of perception, rapid locomotion and receptive movements. Most existing controllers restrict the agility of the robot or are not applicable to interactive sports.
To address these limitations, the researchers developed a system that can integrate perception with the movements of the upper and lower part of the robot body.
They implemented it in a four-legged robot called Anymal-D, which was equipped with a stereo camera for vision-based perception and a dynamic arm to handle a Badminton racket, a task that requires a precise coordination of perception, locomotion and balancing the arms.
Do not miss: More function, less appearance: specialized robots attract greater demand with efficient task management
Robot playing Badminton is a watershed for future control systems
The researchers trained the system based on reinforcement to predict the path of the steering wheel and respond by moving the robot accordingly.
Reinforcement learning is a type of automatic learning in which an agent or system learns to plan effective strategies – to make decisions – based on experimentation with data and interaction with the environment.
In their investigation, Eht Zurich experts tested the robot game against humans and discovered that the machine could move along the track to return blows to different speeds and angles, and that it achieved exchanges of up to 10 consecutive blows.
In addition, the robot could stand on its hind legs to keep the steering wheel in view while preparing to move his arm, but gave priority to his own safety while moving to make sure he did not fall.
The authors suggest that these findings could serve as a basis for future control and perception systems of humanoid robots or robots with legs that need to perform rapid and coordinated movements.
This same Polytechnic School published other “skills” of this type of quadruped robots last year. On that occasion, Anymal proved to be “quite skilled” in ‘Parkour’, a physical activity based on the use of athletic maneuvers to overcome obstacles in an urban environment.
With EFE information
Follow us on Google News to always keep you informed
