Resource optimization algorithms for an automated coordinated CubeSat constellation

Abstract

We present and analyze the performance of two algorithms that plan and coordinate activities for a resource-constrained Earth-observing CubeSat constellation. The first algorithm is the Resource-Aware SmallSat Planner (RASP), which performs low-level planning of observation and communication activities for a single satellite while simultaneously keeping the satellite's onboard resources within specified bounds. RASP utilizes a Mixed Integer Linear Program based formulation and Depth First Search for construction of consistent onboard activity timelines. The second algorithm is the Limited Communication Constellation Coordinator (LCCC), which performs high level coordination of observations across the constellation through a distributed, "weak" consensus mechanism. The performance of the algorithms is tested with a 24 hour simulation of an eighteen satellite constellation over multiple orbital geometries and inter-satellite communication contexts. The orbital geometries include a modified Walker Star constellation and an "ad hoc" constellation defined by historical launches of CubeSats. The multiple communication contexts simulate different methods for sharing observation planning information between the satellites, and include sharing through inter-satellite crosslinks, downlink and uplink to ground stations, connection to a commercial communications constellation, and no sharing at all. Five analyses of the algorithms' performance were conducted, including average revisit times achieved, the numbers of communications links executed, how effectively planning information was shared, the resource margins maintained by the satellites, and the average execution time for the planner. Information sharing significantly aided in balancing revisit times across multiple Earth regions and three sensor choices, reducing the disparity in average revisit times between sensors from 514 minutes to 10 minutes for the Walker case and 617 to 11 minutes for he Ad Hoc case. Significantly more crosslink opportunities were available on average for the Walker satellites than for Ad Hoc (89.2 versus 47.7) and more crosslinks were executed for the Walker case (30.3 versus 20.8). Crosslink was found to be less effective than downlink at sharing planning information across the constellation, with a lower average latency (186 minutes versus 434, Walker) and better average initial timeliness (-35 minutes versus -287, Walker). Information sharing through both a commercial constellation and downlink outperformed sharing through just downlink or just crosslink, with an average latency and initial timeliness of 77 and 74 minutes (Walker). Average data storage and energy storage margins were kept high, as desired, for both constellations, at around 85 and 70 %. RASP planning time was found to scale roughly with the square of planning window length, but stays under a minute in all cases tested (achieving a maximum of 37.71 seconds).