Humanoid robot learns to velvez by reflecting people’s antics

An AI that helps humanoid robots reflect a person’s activities, allowing robots to run, dance and fight in more solid human ways.

The most agile and liquid robotic activities, such as the impressive performance of the robot acrobatics of Boston Dynamics, are usually narrow, pre-orderly sequences. It is still difficult for robots to teach to perform widely performing human activities.

To remove this obstruction, Juanbin Peng At the University of California, San Diego, and their colleagues have developed an artificial intelligence system called exbody2, which allows robots to copy and smoothly execute many different human activities in more vibrant ways.

https://www.youtube.com/watch?v=-QP5UJMRN0

Peng and his team first created a database of actions that could be able to do a humanoid robot, including simple activities, such as standing or walking, more complex maneuvers, such as difficult dance moves. The database included motion captaincy recording of hundreds of human volunteers collected in previous research projects.

Peng says, “Since humanoid robots share a similar anatomical structure with us, it makes sense to take advantage of a large amount of human speed data already available.” “By learning to mimic this type of speed, the robot can immediately understand a variety of human-e-human behaviors. This means that whatever man can do, robots can learn potentially. “

To teach how to walk a simulated humanoid robot, Peng and his team used reinforcement learning, where an AI is given an example of what is involved in a successful movement and then it was assigned to find out How to do it through testing and error. He first learned exbody2 with full access to all data of this virtual robot, such as coordinates of each addition, so that it could imitate human activities as much as possible. Then, they learned it from these activities, but only using data could have access to the real world, such as inertia or motion from the sensor on the body of the real robot.

After being trained on the database, EXBody2 was placed under the control of two separate commercial humanoid robots. It was able to connect simple activities simultaneously, such as walking and bending in a straight line, as well as complicated tricks like following a 40 -second dance routine, punching and walking with a human being.

Peng says, “Humanoid robots do the best when they coordinate all their organs and joints together.” “Hands, feet and torso for many tasks and movements require working together, and the coordination of the entire body significantly increases the range of robot capabilities.”