Plan Recognition in Complex Spatial Domains

We are researching the problem of plan recognition in a complex real-world domain consisting of training battles conducted by actual troops at the US Army’s National Training Center. These battles involve hundreds of participants, last several days, and take place over a very large geographical area. Our task is to identify from this data repeated patterns of movement which correspond to planned "maneuvers" -coordinated activities which generate identifiable patterns of movement that can be identified and used as a basis for prediction. Figure 1 depicts the movements of agents in a training battle over a relatively short amount of time (approximately three hours).