Automatic Generation Of Character Animation For Intelligent Agents Based On Deep Learning

The automatic generation of character animation for intelligent agents is essentially a simulation of life form,which is of great importance to applications ranging from the game industry to special effect production and simulation tests.It not only needs to simulate the intrinsic behavior patterns of the agent(behavior modeling),but also needs to present the external behavior of the agent(character animation),so that the agent can interact with the external environment autonomously and realistically.In this paper,we targeted at the simulation test requirements in autonomous driving task and designed a simulation tool for life form,in which we focused on the automatic generation of character animation for both biped characters and quadruped characters.Our work can be summarized as follows:1)Simulation of the intrinsic behavior pattern of agents(behavior modeling).We modeled characters as agents,which means the character can perceive the environmental information,plan a path to the destination and make behavioral decisions.2)Presentation of the external behavior of agents(character animation).For biped characters and quadruped characters,different solutions based on deep learning are proposed respectively,so that characters can make coherent,realistic and diverse actions in real time according to the terrain.3)Solution and prototype system for automatic generation of character animation for agents.We integrated the above technologies to form the final solution and implemented a prototype system.Several representative animations have been implemented to test our solution.