论文部分内容阅读
结合前瞻搜索思想提出了一种判断模型近似行为等价的方法,首先通过比较候选模型的部分解(即策略树)判断模型近似行为等价,然后自上而下对近似行为等价模型进行快速聚类和修剪,利用代表模型将交互式动态影响图扩展成为平铺动态影响图,最后求解平铺动态影响图.算法减少了候选模型的存储空间和运行时间,提高了算法的效率.最后通过多Agent老虎问题及音乐会问题的实验验证了该方法的有效性.
In this paper, a method to judge the approximate behavior of the model is proposed by combining the look-ahead search idea. First, the approximate behavior equivalence is judged by comparing the partial solutions of the candidate models (that is, the strategy tree), and then the near- Clustering and pruning, using the representative model to expand the interactive dynamic impact graph into a tiled dynamic impact graph, and finally solving the tiled dynamic impact graph.The algorithm reduces the storage space and running time of the candidate model and improves the efficiency of the algorithm.Finally, Experiments with multi-agent tigers and concerts demonstrate the effectiveness of this method.