The line between robotic and human movements is becoming increasingly blurred. At the University of California, San Diego, a team of researchers has developed an AI capable of teaching robots to waltz (and much more), opening up new perspectives in mass robotics.
The Dance of Robots: From Waltz to Combat
Until now, the most fluid movements of robots, such as the spectacular acrobatics of Boston Dynamics, were pre-programmed and limited sequences. But Xuanbin Peng and his team ofUniversity of California, San Diego they are changing the rules of the game with ExBody2, an artificial intelligence system that allows robots to copy and replicate human movements in a natural way. Meanwhile I'll link the search here, then I'll tell you about it.
ExBody2’s innovation isn’t limited to dance, of course. The system allows robots to perform a wide range of movements, from simple walking to complex waltz sequences and even combat moves. The key to success? An approach that exploits the similarity between the physical structure of humanoid robots and that of the human body.
Since humanoid robots share a similar physical structure to us, it makes sense to take advantage of the huge amount of human movement data already available. By learning to mimic this type of movement, the robot can quickly acquire a wide range of human-like behaviors. This means that anything humans can do, the robot can learn to do.
The movement database
The team created a comprehensive database of actions that a humanoid robot could perform. This archive includes motion capture recordings of hundreds of human volunteers, collected in previous research projects. This is the data that formed the basis for the robot's learning.
Reinforcement learning
To teach a simulated humanoid robot how to move, the researchers used reinforcement learning. In this process, the AI is given an example of a correct movement and must then learn to replicate it through trial and error. The process took place in two stages: first with full access to the virtual robot's data, then using only data that would be available in the real world.
Virtual waltz, real waltz
After training on the database, ExBody2 was tested on two different commercial humanoid robots. The results were surprising: the system was able to fluidly combine simple movements such as walking in a straight line and crouching, but also to perform more complex moves such as a 40-second dance routine, throwing punches and, of course, waltzing with a human partner. The plus acquired? Coordination. As he explains Peng:
Humanoid robots work best when they coordinate all their limbs and joints together. Many tasks and movements require the arms, legs, and torso to work in sync, and full-body coordination greatly expands the robot's capabilities.
What does this development entail?
The study by the University of San Diego opens new perspectives in the field of humanoid robotics and of the embodiment in general. The ability to learn and replicate complex human movements is not only an interesting technological demonstration, but could have practical applications in many areas, from healthcare to entertainment.
Now the dancing gets serious: the waltz is just the beginning. The technology developed by Peng and his team could be applied to any type of human movement. In the near future, we will see robots that move in increasingly natural and fluid ways in all types of situations, increasingly capable of interacting naturally with our world.