Generació d’animacions de personatges de jocs de lluita mitjançant xarxes neuronals

Abstract

L’objectiu principal d’aquest treball és crear un escenari/prototip on hi hagi un personatge que estigui animat amb algun mètode procedural i crear una intel·ligència arti_cial que li proporcioni les dades necessàries. Per ser més concrets, la idea és crear un escenari on un personatge ataqui a un altre amb una espasa i el defensor haurà de defensar-se movent la seva espasa de forma procedural. Per dur-ho a terme s’ha considerat dos elements: Cinemàtica inversa i xarxes neuronals. Per saber on mourà l’espasa defensora, es fa servir una intel·ligència arti_cial formada per un model de xarxa neuronal, la qual rep dades de l’atacant i algunes dades del defensor, les processa i dona la posició on hauria d’anar l’espasa per defensar-se de l’atac. Llavors, fent servir cinemàtica inversa (Inverse Kinematics), es mou els braços del defensor tot creant una animació procedural de moviment de braços. Sabent prèviament la complexitat de la cinemàtica inversa i les xarxes neuronals, s’espera com a resultat una animació simple, és a dir, en cap moment es busca crear una animació complexa i espectacular capaç d’igualar el treball d’un animador professional.

The main objective of this work is to create a scenario/prototype in which there is a character that is animated with some procedural method and to create an arti_cial intelligence that will character that is animated with some procedural method and create an arti_cial intelligence that provides the necessary data. the necessary data. To be more specific, the idea is to create a scenario in which one character attacks another character with a sword and character attacks another character with a sword and the defender must defend himself by moving his sword in a procedural way. For this purpose, two elements have been considered: Inverse kinematics and neural networks. To know where the defender's sword will move, we use an artificial intelligence formed by a neural network model, the a neural network model, which receives data from the attacker and some data from the defender, processes them and gives the position where it should move. processes it and gives the position where the sword should go to defend against the attack. Then, by making Inverse Kinematics, the defender's arms are moved, creating a procedural animation of arm movement. procedural animation of arm movement. Knowing previously the complexity of the inverse kinematics and neural networks, a simple animation is expected as a result, i.e., at no time is it intended to create a complex and spectacular animation capable of matching the work of a professional animator.