Kento Tomita, Georgia Institute of Technology; Yuri Shimane, Georgia Institute of Technology; Koki Ho, Georgia Institute of Technology
Keywords: Cislunar Space Situational Awareness, Predictive Sensor Tasking, Demand Uncertainty
Abstract:
Space Domain Awareness (SDA) requires systems adaptive to its dynamic environment. The demands of information collection evolve over time, which should be incorporated into the SDA systems architecture design. For example, the translunar and cislunar regions will see both stable observations demands as well as potentially unanticipated detection and tracking demands. The former type of demands involves monitoring and maintaining a custody chain of existing assets, such as the Gateway from the Artemis program, LunaNet service provider satellites, CLPS satellites loitering at low-lunar orbit, as well as numerous other space lunar missions by both US and non-US entities, such as CNSA’s Queqiao relay satellite at Earth-Moon L2. The latter type comes primarily from translunar transfers, both heading to and returning from cislunar space. Missions with mothership-daughtership configurations, such as JAXA’s SELENE mission launched in 2007, may also deploy additional flying segments after reaching their destination, leading to additional targets to be monitored. The cislunar SDA systems architecture should be designed to meet these dynamic demands.
To design such a demand-adaptive Cislunar SDA architecture, examining how an SDA system would react to a new demand is critical. Although we are gaining knowledge about the SDA performance for different system architectures, such as how many observer spacecraft in which trajectories, most of the existing work assumes a fixed set of target objects or areas of interest; the system’s performance against more dynamic observation demands, such as a sudden increase of the number of targets, have yet to be discussed extensively. Analyzing the system’s capability for new, potentially unanticipated demands is crucial because lacking this analysis may cause a lost target; the uncertainty projected onto the sensor frame could be sufficiently larger than the sensor field-of-view, resulting in unnecessarily increased demand for the initial survey. In the cislunar regime, this is even more pronounced as the dynamics is more chaotic due to strong perturbations coming from multiple effects, including but not limited to the Earth and Sun’s third-body effects as well as the complex gravitational field of the Moon. Since frequent data acquisition of these targets is essential for tracking and orbit determination, the resource allocation and margin management should take into account such dynamic demands.
As a building block of the system’s performance evaluation against the demand uncertainty, we study the predictive sensor tasking algorithm via integer linear programming (ILP) with the rolling horizon method. It should be noted that a variety of sensor tasking algorithms for catalog maintenance have been studied: metric-based reactive methods with the Fisher information, entropic measures, and Lyapunov exponents, optimization-based predictive methods such as gradient-based stochastic optimization, dynamic programming, reinforcement learning, or Monte Carlo Tree Search (MCTS). We use ILP with rolling horizon methods mainly to benefit from its formality and efficiency. Since we are interested in the effect of prior knowledge about the future demand on the system’s performance to examine how effective or not to use predictive tasking control instead of reactive control.
We evaluate the SDA performance for several different sets of future demand knowledge and the target and observer trajectories. With the summary of the system’s performance for a set of SDA architectures and the reaction to increased demands, we aim to offer building blocks for more complex campaign-level optimization problems.
Date of Conference: September 19-22, 2023
Track: Cislunar SDA