Representation and Recognition of

In this work we extend from the single agent to the on-line multi-agent stochastic policy recognition problem using a network structure. By using knowledge of agents’ interrelations we can create a policy structure that is compatible with that of a hostile military organisation. Using this approach we make use of existing knowledge about the military organisation and thereby strongly reduce the size of the hypothesis space. In this way we are able to bring down the problem complexity to a level that is tractable. Also, by using statistical models in policy recognition we are able to deal with uncertainty in a consistent way. This means that we have achieved improved policy recognition robustness. We have developed a proof of concept Bayesian Network model. For the information fusion purpose, we show with our model that it is possible to integrate the pre-processed uncertain dynamical sensor data such as the enemy position and combine this knowledge with terrain data and uncertain a priori knowledge such as the doctrine knowledge to infer multi-agent policy in a robust and statistically sound manner. 1.0 INTRODUCTION Dealing with uncertain information in a complex and in some cases chaotic environment is the difficult task for military commanders. A set of methods which improve the process of collecting and reasoning about uncertain information is called information fusion. The goal of information fusion is to describe a particular state of the world of interest by making best possible use of all available information. In military applications this can apply to anything from the position and type of hostile forces to an enemy’s plans and intentions. This paper shows how our knowledge about the enemy can be represented and his policies recognised. We choose Bayesian Networks (BN) as the method for representation of our knowledge about the enemy, designated static knowledge, and Dynamic Bayesian Networks (DBN) for inference based on sensor data, designated dynamic knowledge. Those problems, as many others, must of course be addressed in the operational system. In order to focus on policy recognition, in this work we do not deal with the classical identification and association problems. The stochastic nature of policies is derived from the fact that we do not have full knowledge about the enemy and his actions. Instead, policies are represented as discrete probability density functions on different modelling levels. This means that military commanders should not only pay attention to the policy with the highest probability but to all policies that have a significant probability to occur. Paper presented at the RTO IST Symposium on “Military Data and Information Fusion”, held in Prague, Czech Republic, 20-22 October 2003, and published in RTO-MP-IST-040. RTO-MP-IST-040 10 1 Report Documentation Page Form Approved OMB No. 0704-0188 Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information, including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it does not display a currently valid OMB control number. 1. REPORT DATE 00 MAR 2004 2. REPORT TYPE N/A 3. DATES COVERED 4. TITLE AND SUBTITLE Representation and Recognition of Uncertain Enemy Policies Using Statistical Models 5a. CONTRACT NUMBER