Session
Technical Session 12: Constellation Missions
Location
Utah State University, Logan, UT
Abstract
Large constellations are quickly becoming the norm in small satellite missions. These constellations are being designed and developed faster than ever before through the utilization of smaller, heterogeneous spacecraft. Often, these constellations provide increased resiliency and capabilities over their heritage, highly tailored counterparts. The ability to replace on-orbit assets quickly and with lower costs is an advantageous feature of these large smallsat constellations. With the advent of these new architectures, though, come increased complexity in mission operations. The management and monitoring of potentially hundreds of heterogeneous space assets can be extremely challenging and negate much of the cost savings using current operational approaches. Additional complexity is added with the loss expectancy of some number of assets inherent to the design within these constellations. Rather than tasking individual assets to complete missions on behalf of the system, ideal operation would be conducted through tasking of the constellation as a whole. This approach requires tasking of the individual assets by the constellation using machine-to-machine (M2M) data sharing and on-orbit autonomous decision making. Recent advances in machine learning (ML) and artificial intelligence (AI) have now set the stage for the state of the possible in this regard.
The authors of this paper are part of a research and development team aiming to develop solutions and tools to support this operational approach. The ideas presented involve a procedural and technical implementation of using forecasted operational effects developed by a combination of state machines and ML tools. First, the system’s state is gathered, time-synced, and produced into a “Dynamic Relative Telemetry Calculator.” This is presented as an NxN matrix documenting each node’s state relative to all other nodes in the system. Next, a desired operational command can be loaded into the system. Multiple possible operational scenarios and effects can be propagated. For each propagation, each asset must be capable of reporting the “cost” of performing a certain task within a certain operational scenario. By itself, this still requires a human in the loop to analyze the results and determine a command decision. However, the secondary and tertiary effects of these decisions are still unknown. To this front, the authors are developing a method of wrapping ML capability around the system's state machine and propagators to create a forecaster capable of autonomously determining optimal decisions within a system. The forecaster operates in real time, improving its predictions as more data is produced by each subsystem. Generated operational forecasts, and their effects, are validated with log data from a simulation. This data is being applied to proprietary mission scenarios, but could also be applied to historical open/mission data for validation or operational lessons learned. Over time, this forecasting tool could optimize large constellation management by reserving human in the loop for only the most severe/impactful decision thresholds. This paper will present current progress of the integrated solution, next steps in the research and development roadmap, and, most importantly, the current technical hurdles still to overcome to achieve true spaceflight autonomy.
Constellation Forecasting Tools for Autonomous Operations
Utah State University, Logan, UT
Large constellations are quickly becoming the norm in small satellite missions. These constellations are being designed and developed faster than ever before through the utilization of smaller, heterogeneous spacecraft. Often, these constellations provide increased resiliency and capabilities over their heritage, highly tailored counterparts. The ability to replace on-orbit assets quickly and with lower costs is an advantageous feature of these large smallsat constellations. With the advent of these new architectures, though, come increased complexity in mission operations. The management and monitoring of potentially hundreds of heterogeneous space assets can be extremely challenging and negate much of the cost savings using current operational approaches. Additional complexity is added with the loss expectancy of some number of assets inherent to the design within these constellations. Rather than tasking individual assets to complete missions on behalf of the system, ideal operation would be conducted through tasking of the constellation as a whole. This approach requires tasking of the individual assets by the constellation using machine-to-machine (M2M) data sharing and on-orbit autonomous decision making. Recent advances in machine learning (ML) and artificial intelligence (AI) have now set the stage for the state of the possible in this regard.
The authors of this paper are part of a research and development team aiming to develop solutions and tools to support this operational approach. The ideas presented involve a procedural and technical implementation of using forecasted operational effects developed by a combination of state machines and ML tools. First, the system’s state is gathered, time-synced, and produced into a “Dynamic Relative Telemetry Calculator.” This is presented as an NxN matrix documenting each node’s state relative to all other nodes in the system. Next, a desired operational command can be loaded into the system. Multiple possible operational scenarios and effects can be propagated. For each propagation, each asset must be capable of reporting the “cost” of performing a certain task within a certain operational scenario. By itself, this still requires a human in the loop to analyze the results and determine a command decision. However, the secondary and tertiary effects of these decisions are still unknown. To this front, the authors are developing a method of wrapping ML capability around the system's state machine and propagators to create a forecaster capable of autonomously determining optimal decisions within a system. The forecaster operates in real time, improving its predictions as more data is produced by each subsystem. Generated operational forecasts, and their effects, are validated with log data from a simulation. This data is being applied to proprietary mission scenarios, but could also be applied to historical open/mission data for validation or operational lessons learned. Over time, this forecasting tool could optimize large constellation management by reserving human in the loop for only the most severe/impactful decision thresholds. This paper will present current progress of the integrated solution, next steps in the research and development roadmap, and, most importantly, the current technical hurdles still to overcome to achieve true spaceflight autonomy.
