Comtuter and Internet News Channel
Researchers at the University of California at Berkeley and the University of British Columbia have created virtual models that can train in melee, parkour and acrobatics, honing their movements to perfection, just like a man. Their work can significantly affect the creation of computer games and films.
“The actor can show the machine some examples of tricks, and the system uses them in different situations,” explained Jason Peng, one of the authors of the study.
One algorithm can be used to train a variety of complex tricks. Image: Berkeley Artificial Intelligence Research
Virtual characters, which scientists created, use an algorithm of stimulated learning, based in part on methods of teaching animals.
Researchers recorded movements of professional acrobats and experts in the field of martial arts. The virtual model tries to repeat these movements and each time gets approval if it starts to work out. To use this technique it is necessary that the character has a realistic body and is in the world with the correct laws of physics.
With the help of the algorithm, you can teach the hero to do a somersault or a lunar walk. “In this way, you can solve a lot of animation problems,” said Sergei Levin, an associate professor at the University of California at Berkeley, who also took part in the project.
Characters of high-budget computer games can look very realistic, but in fact they are just digital puppets that act clearly under the script. In geymdev and animation actively study the use of programs that automatically add realistic physics to the characters. James Jacob, the head of Ziva Dynamics, believes that stimulating learning will add realism not only to the character’s appearance, but also to his behavior. “Before, people had to use more simple methods,” he said. “Now we are teaching the digital model to understand the movements of a person or a being. We direct it and teach it to adapt to the environment. “
Tricks are not given immediately. Image: Berkeley Artificial Intelligence Research
This method can be useful not only in the film and games industry – with the help of simulation tricks can learn and real robots. For example, you can first teach the robot to collect a table in a virtual space, so that later he tries his hand in the real world.
Levin believes that ultimately robots can even teach people new tricks. “If someone wants to do a trick that no one has ever tried, you can try it out in the virtual space and, maybe, in the end, something interesting will turn out,” the scientist commented.
Materials on the topic: