Artificial intelligence and neural networks with the function of machine learning have long been able to learn various skills with little or no help. But, according to Techxplore, a group of researchers came up with the idea of making AI act as a "mentor", teaching computer models to run, walk, play football and overcome obstacles based on their own experience.
The new algorithm is called DeepLoco, and it can be used as an alternative way of creating motion animations in games and movies. Animation of computer characters now happens manually, or through the transfer of pre-recorded movements of real actors. DeepLoco also allows computer characters to automatically move so that the movements are realistic and consistent with the environment. During the training process, the characters get experience through trial and error, developing an ideal behavior model for each specific situation.
For example, already now the characters have learned to walk on a narrow path, avoid collisions with moving objects and even kick a soccer ball. In this case, according to the creators of the algorithm, the learning of computer models is not limited to. DeepLoco can also be used in training two- and four-legged robots. The application of the process of "independent learning" can give an opportunity to teach robots movement without precise manual adjustment. As the author of the work, Professor Michel van de Pann,
"We create models of people who learn to move quickly and dexterously, focusing on the environment. We teach computer characters to react to the situation without having to manually indicate the necessary strategies, for example, how to balance or plan the path through moving obstacles. All this model learns independently. "
The article is based on materials .
Comments
Post a Comment