Pythia Phase II SBIR

Stellar Science proposes Pythia, a distributed multitarget manager as a proliferated low-Earth orbit (pLEO) expert system for autonomous target-track management. Each node in the Pythia distributed architecture is comprised of a multitarget track manager as well as a sensor task manager. Each track manager employs the sample greedy gossip information consensus filter (SGG-ICF) protocol to autonomously collaborate with neighboring managers and converge on global track-state estimates, while maintaining a low communication burden. Each distributed task manager utilizes the consensus bundle based (CBB) algorithm to reach task allocation consensus based on hypothesized gain of information value. Our Phase I effort integrated the Pythia expert system for pLEO autonomous target track management into an existing government-owned modeling, simulation, and analysis (MS&A) tool, the Advanced Framework for Simulation, Integration, and Modeling (AFSIM). This MS&A framework has enabled us to test our expert systems for autonomous target track management against scriptable, multi-domain, physics-based scenarios, successfully demonstrating the feasibility of the integrated system as a digital twin, ideal for expert system verification and validation, constellation design studies, and trade-space analysis of planned architectures, as well as inclusion in tabletop exercises, simulated wargames, or other operational-like environments. In Phase II, we propose to build upon our Phase I work and improve data and track association, network discovery method, and communication management within the distributed architecture. We plan to complete and upgrade the implementation of the simulation environment started in Phase I, perform parameter studies to optimize the parameters employed by our algorithm implementations, as well as optimize constellation and satellite designs. We will continue to build trustworthiness of our autonomous target track management system by testing against well-established baselines. In addition to full modularization of our algorithm implementations in C++, such that they are fully portable beyond our specific simulation environment, we will develop Python interfaces to these implementations to maximize their flexibility in use. We will also continue to develop plug-ins for the above algorithms for their use in AFSIM. Our Phase II simulation effort will involve a more realistic implementation of system constraints, including fully perception-based data and track association algorithms, communication management, processing latencies, etc., to better afford an analysis of the challenges and impacts of deploying the Pythia expert system on-orbit in low-to-medium size, weight, and power (SWaP) satellites. We plan to leverage the mature simulation builder tool, Galaxy, to efficiently manage the needed workflows for a multirun environment to enable broad trade-study analyses that can inform the real-world constellation design constraints and paradigms.