Multi-agent Cooperation Method Based on Intention Recognition

The paper discusses the cooperation method among multiple agents in 3D space based on intention recognition. It is assumed that an observer observes multiple moving objects that are its company and the enemy. The purpose of our work is to help the observer to make cooperation decision efficiently. Due to the distortions the perspective brings, cooperating in 3D space is difficult. To contact agents with each other, we list two kinds of tracking model to analyse the problem, for the observer's teammates can estimate the enemy's trace to adjust their flight paths, and also can just move towards the directions of the enemy. The relationship among the agents and the moving intentions of them are to be estimated by sequence images. Aimed at each tracking model, estimation algorithms are proposed respectively by which the probabilities of intention are gained. First, we analyse the characteristic of the two models in detail. Then an estimation function is listed to predict the possibility of tracking trends and it is proved that there is also a normal distribution relationship between agents with the Kalman filter. The proposed approaches are applied to process several groups of simulated images at the end. The result of the experiments proves the availability and the robustness of the proposed method.