Collaborative Human-like Agent Based Virtual Rescue Training System

Virtual Training system is a virtual environment for training, which is implememted by the virtual reality technology. For more lifelike imitation and good training effect, it crys for the virtual world to add lifelike characters (virtual human).Since the Agent has the independence, autonomy and intelligence characters, it is convenient to construct Agent-based intelligence model to simulate human to construct a high realistic virtual training environment by Agent technology. In the virtual environment, the lifelike characters (virtual human) and the user Avatars are implemented by intelligentAgent—human like Agent. In the virtual rescue training system, there ismulti-human collaboration, such as single human cannot accomplish some tasks, and he needs help; or he can accomplish the task by himself, but he needs to know some information depending on others to accomplish the task. All these need to fetch the Multi-agent collaboration technology in the virtual training system. After reading up a great deal of literature about the application of intelligent Agent and the Multi-agent in virtual environment, for the insufficiency of these research works, the dissertation focuses on the virtual human modeling and interaction and collaboration between the human like Agents in Virtual Environment, including the design of the virtual human's model, the interactive behaviors of human like Agents for the virtual environment, the collaboration work of Multi-agent and the corresponding model.On the virtual human modeling, firstly, the structure of the virtual human is discussed in the thesis, the sense module, thinking module, plan module and behavior module are included. These modules are integrated into a system. Secondly, the dissertation puts forward an implement frame to construct a virtual human who has vivid human form and general intelligence (such as percept, belief, plan, action etc). According to the plan and behavior capability of the virtual human, virtual human's actions were divided into atom actions, and these actions were realized by H-Anim parametric animation.On the multi human-like Agents interaction for virtual environment, firstly, this thesis proposed the interactive hierarchy of Agents in virtual environment, and designed the interactive algorithm based on this frame. Secondly, in order to solve the conflicts between Agents, the nearest neighbor algorithm to match the instance of Agents was adopted. Thirdly, the thesis researches the model for virtual Agents group, implemented an Agents group which have the most similarity. Finally, we presented a Multi-agent collaborate model, represented the Agents team work process state using the Petri Net to solve the parallel and synchronization action.The dissertation also introduced the pro-active information exchange and dynamic role selection to advance system efficiency and flexibility.Eventually, according to the detail requirement of the earthquake incident rescue training, the dissertation gave a design of virtual training system for earthquake incidence rescue drilling, and implemented it.