Description:
TECHNOLOGY AREA(S): Info Systems
OBJECTIVE: Develop robust, near real-time algorithms that rapidly discover the behavioral patterns and operational intent of potentially evasive and/or ambiguous active resident space objects (RSOs) for the purposes of space situational awareness (SSA) across the entire SSA space catalog.
DESCRIPTION: Space protection and SSA require rapid and accurate space object behavioral and operational intent discovery. Ground- and space-based Air Force surveillance assets are a critical foundation of U.S. space control operations. Optimally and autonomously controlling their actions effectively and in real-time is fundamental to space object evasive and ambiguous behavioral pattern identification. The problem of behaviorally evasive intent identification is challenging for several reasons. For example, surveillance assets do not completely observe all RSO variables, and system and subsystem parameters required to infer intent. RSOs can potentially be reactive, continuously responding to their perceived environment and choosing their actions correspondingly in order to evade discovery of their capabilities. The problem is further complicated by the fact that the process of intent and capability discovery is fraught with uncertainty in the underlying behavioral pattern models and RSO states, in the observation process, and in the behavioral policy pursued by the RSO. Finally, it is also desired to select a set of surveillance actions that maximize the likelihood of behavioral and capability identification. Due to the large number of hypothetical actions, counter-actions and counter-counter-actions made by the surveillance asset and the RSO over a future look-ahead window of time, along with the large number of RSOs in the space catalog, the problem of optimizing the surveillance asset’s actions over the look-ahead period is computationally intractable. Given a surveillance asset’s capabilities, the ability to identify the existence of undiscoverable RSO “blind-spot” behaviors is critical. Advanced algorithms to process a diverse set of raw sensor data and optimal action selection under uncertainty for enhanced behavioral intent and capabilities discovery are needed. Such algorithms must be highly responsive and adaptive despite the curse of dimensionality that underlies the optimal operational intent identification problem. Existing and new reliable RSO probabilistic patterns of behavioral models need to be utilized. Such models describe the set of possible states an RSO may assume and how these states can transition from one to the other given a surveillance asset’s chosen action. An appropriate utility function for the optimal surveillance policy needs to be developed. Such a function should be designed in order to discover an RSO’s intent, if possible, in the shortest amount of time with the highest level of confidence level given the uncertainty underlying the problem. This topic solicitation addresses the problem of behavioral intent and operational capability discovery within an uncertain game theoretic context, with an interest in improved optimal surveillance asset action selection for rapid identification. Innovative solutions are sought for efficient and rapid discovery despite behavior model uncertainty and RSO action strategy under potentially large number of evasive strategies. Algorithms that are capable of processing raw observation data along with intelligence data and environmental data will be of particular interest.
PHASE I: Develop the mathematical basis for dynamic behavior models to enable near real-time behavioral patterns and operational intent identification. Develop algorithms that compute surveillance asset optimal policies under modeling uncertainty. Identify techniques to detect technological gaps in identifying intent under evasive and/or ambiguous active RSO behavior. Provide a prototype demonstration as applied to a few RSOs in the catalog.
PHASE II: Provide a scalable prototype demonstration of the technology in a realistic environment using realistic data with errors and biases as well as realistic processing speeds in complex scenarios. Extend algorithms to accommodate different sensor designs and sensing environments. Demonstrate scalability with respect to behavioral patterns model parameter space, as well as with respect to the number of RSOs. The solution must be demonstrably scalable over the number of RSOs in a representative SSA space catalog.
PHASE III: Rapid evasive intent and behavioral identification under uncertainty is highly applicable to many military (Joint Space Operations Center), civilian and public safety and security uses.
REFERENCES:
1: Kaelbling, L.P., Littman, M.L., Cassandra, A.R. "Planning and acting in partially observable stochastic domains". Artificial Intelligence Journal, Vol. 101: pp. 99–134, 1998.
2: Mertens, J. F. & Neyman, A. "Stochastic Games". International Journal of Game Theory Vol. 10, No. 2, pp. 53–66, 1981.
3: Neyman, A. & Sorin, S. "Stochastic Games and Applications". Dordrecht: Kluwer Academic Press, 2003.
KEYWORDS: Satellite Characterization, Behavior Modeling, Dynamic Behavior Models, Sensor Optimization
