The Berkeley Hybrid Robotics Laboratory (University of California) has published a study on building a dynamic robotic doorman with machine learning. Capable of saving 87.5% of shots on goal,
Where the nine kg quadruped robot is called mini cheetah, Involves dynamic movement To manipulate objects, move fast in any direction and combine a planner to plot the trajectory of the end effector.
Skills acquired are dodging interceptions, diving to reach the bottom/top corners, and jumping to cover the top of the target. Furthermore, the idea is that all the mentioned actions are recoverable and that Robot lands safely on its feet,
mini cheetah programmed manually And your system is trained in simulation. In the following video, you can see the android defending a goal that is 1.5 meters wide and 0.9 meters high with a ball that is kicked from four meters away.
The researchers note in the study that “our system can be used to directly transfer the dynamic maneuvers and goalkeeper skills learned in the simulation to a real quadrupedal robot, with a successful intercept rate of 87.5% of random shots in the real world.” Hopefully this document brings us a step forward towards allowing robotic players Competition with humans in the near future,
Sign up for our newsletter and receive the latest tech news in your email.