W. Thomas Vestrand, Los Alamos National Laboratory; Przemek Wozniak, Los Alamos National Laboratory; Sean Brennan, Los Alamos National Laboratory; Troy McVay, Los Alamos National Laboratory; Lucas Parker, Los Alamos National Laboratory; Rebecca Holmes Sandoval, Los Alamos National Laboratory; Yancey Sechrest, Los Alamos National Laboratory
Keywords: Space Traffic Management, Space Domain Awareness, Artificial Intelligence
Abstract:
Economic opportunities and new technology, which is lowering the barrier of entry for access to near earth orbit, are generating an accelerating pace of space activity and a rapidly expanding, global, commercial space industry. The task of monitoring and managing this international space traffic is now threatening to overwhelm traditional approaches and legacy systems. Modernization of real-time Space Situational Awareness (SSA) capabilities and the establishment of new Space Traffic Management (STM) norms and protocols are therefore becoming essential for both global security and future economic vitality of the space enterprise.
Central to the improvement of SSA and STM capabilities is the development of new approaches that enable the collection of more timely, comprehensive and reliable space surveillance observations. Timely collection poses the challenge of coordinating sensors that span a global footprint and is likely to require international coordination. So how does one build an architecture that enables real-time space surveillance data collection by independent agencies/organizations with different priorities while ensuring that the integrated, autonomous, collection provides reliable measurements? Here we present simulations of a collaborative STM architecture concept, called a Dynamic Coalition Architecture (DCA), that is designed to address this real-time coordination challenge.
The Dynamic Coalition Architecture employs Artificial Intelligence (AI) methods drawn from distributed problem solving and multi-agent system approaches to organize sensors controlled by independent organizations into an efficient, autonomous, collection ecosystem. One of the key features of the DCA is that membership in each collection coalition is negotiated in real-time through the use of software agents that are configured with coalition acceptance criteria by the sensor owners. This conservation of local control helps solve problems associated with organizational/security boundaries and ensures sensor compliance, at any given time, with the priorities of the owner.
The autonomous nature of the DCA approach make it especially powerful for space traffic management where anomalies often have to be quickly followed up and resolved in real time. As each potential traffic anomaly is recognized, the DCA generates follow-up coalition invitations in real time to ecosystem sensor agents. Some of the key information included in the coalition participation request is a characterization of the anomaly, a measure of the confidence for the anomaly, and the spatial and temporal window of opportunity for collection by the sensor agent. Once coalition membership is negotiated, a coalition is spawned and participating agents are assigned roles that reflect their capabilities. The assignment of coalition roles in the autonomous collection ecosystem is based on meta-data about the collection agents and the appropriate measurements as determined by the AI cognition/tasking agent. By matching the collection event type with a scheme for optimal collection, the tasking agent works to make the right collection, at the right time and at the right place to collect the observations that are needed to resolve the anomaly.
An important aspect of ecosystem collection optimization is to efficiently match the sensor agent capability to the collection coalition needs. The agents in most sensor ecosystems are heterogeneous and even those agents with similar capabilities are not all equally effective. Key effectiveness metrics that help shape the optimization include: Measurement fidelityAre the sensor agent measurements reliable and, when available, consistent with test calibrators?; AccessibilityHow often is the agent on-line and responding in a timely manner?; DependabilityHow often does the agent successfully collect when assigned a role in the coalition?; and AbilityIs the agent capable of conducting the measurement that is needed? The DCA tasking agent tracks agent effectiveness, as a function of task type, to inform future collection coalitions and to help optimally allocate ecosystem resources.
We present simulations of an AI-enabled, global surveillance ecosystem that applies the DCA approach to real-time space traffic management. The simulations utilize an AI cognition/tasking agent that employs Case-Based Reasoning (CBR) techniques and agent effectiveness metrics to allocate ecosystem resources. In CBR, each potential collection coalition is mapped to an efficient execution plan, called a case, which describes what measurements need to be made and when they need to be made to collect the key information. The initial case recipes are generated based on: (1) historical responses that were successful and (2) new approaches developed by subject matter experts. CBR employs four main steps: retrieve, reuse, revise and retain to boot-straps these cases and spawn new cases as the system learns. This provides a mechanism to quickly improve the data collection solutions and allows the system to learn by retaining successful responses as new cases. It also provides an easy pathway for the virtual system to bootstrap performance improvements from test scenarios. We discuss how these simulations demonstrate the challenges associated with real-time, autonomous, space surveillance and allow us to start to quantifying the utility of AI driven anomaly follow-up, negotiation, and sensor trust modelling that the DCA approach enables.
Date of Conference: September 14-17, 2021
Track: SSA/SDA