Plan detection under partially observable and cluttered conditions

This paper examines the problem of detecting the execution of plans performed in partially observable and cluttered environments. In particular, a plan is defined as a series of tasks that must be executed according to a known precedence relation and build toward some final objective. The goal of a plan detection algorithm is to detect the execution of the plan from the available ambiguous and incomplete data before it reaches the terminal event. This paper presents a Monte Carlo inference algorithm capable of estimating the belief that the plan is currently being executed and how much progress has been made. The performance characteristics of the algorithm are tested for a variety of simulated data sets containing different signal to noise ratios.