Allan Shtofenmakher, Massachusetts Institute of Technology; Hamsa Balakrishnan, Massachusetts Institute of Technology
Keywords: Space-Based Space Situational Awareness, Space-Based Optical Sensors, Sensor Tasking, Sensor Scheduling, Space Domain Awareness (SDA)
Abstract:
As the population of objects in low Earth orbit (LEO) continues to skyrocket, the operational systems and algorithms used to prevent in-space collisions and preserve space sustainability must be improved. At present, the U.S. Space Surveillance Network (SSN) maintains a catalog of over 20,000 resident space objects (RSOs) in LEO, tracked primarily by ground-based sensors [1]. A key weakness of ground-based space situational awareness (GB-SSA) systems is their limited geographic distribution, which leaves significant gaps in SSN coverage over portions of Europe, East Asia, and the Pacific and Atlantic Oceans [2]. By contrast, space-based sensors, with the exception of those in geosynchronous Earth orbit (GEO), are not limited to any particular longitudes.
Unlike their ground-based counterparts, space-based space situational awareness (SB-SSA) sensors are generally unaffected by atmospheric and weather conditions [2,3]. They offer more dynamic fields of regard than ground-based systems by virtue of their perpetual motion about the Earth [4]. SB-SSA sensors tend to be more sensitive than GB-SSA ones and, consequently, can detect smaller and dimmer RSOs [5]. The Space-Based Visible (SBV) sensor onboard the Midcourse Space Experiment (MSX) satellite first demonstrated the potential of using onboard satellite sensors to detect and track RSOs in 1998 [6]. Since then, space-based space surveillance has been integral to the SSN’s mission, with additional space-based assets deployed every few years [7,8,9,10,11,12]. All such assets have been optical in nature [2], although recent work has explored the feasibility of using radar [13,14,15] and laser-based [16,17,18] systems for SB-SSA.
Despite their many advantages, the deployment of dedicated SB-SSA systems has been quite limited [5]. Four NorthStar-1 spacecraft, launched in January 2024 [19], constitute the first commercial satellite constellation dedicated to SB-SSA [20]. There is currently no known U.S. governmental satellite constellation dedicated to supporting the LEO space situational awareness (SSA) mission. Although the Space-Based Space Surveillance (SBSS) program was originally intended to include a constellation of four spacecraft at full deployment, as of January 2025, only one such $850M spacecraft has been launched [2,8,21]. Nevertheless, both the United States and Europe are considering constellations of SSA spacecraft that would be capable of performing LEO-to-LEO SSA for the purposes of catalog maintenance [21,22].
Limited published research has focused on the tasking and scheduling algorithms used to assign space-based assets to track LEO RSOs. Miller’s special perturbations (SP) Tasker [23], the current operational algorithm for tasking (but not scheduling) the few existing SB-SSA systems to track objects primarily beyond LEO [2,9,10,21], became operational in 2005 and has struggled to keep pace with the dramatic population growth in Earth orbit. Recently published work to improve tasking or scheduling of multiple space-based sensors for SSA has typically implemented uncoordinated, single-sensor optimization strategies. For example, Paul and Lee [24] used mutual information gain methods to task and schedule a single simulated ground-based optical telescope being supported by 12 space-based sensors, each of which passively collects data on 21 incidentally observed LEO RSOs. Similarly, Roberts et al. [25] used deep reinforcement learning to train a single space-based optical observer in LEO to track up to 400 near-GEO RSOs over a 60-min interval. To extend this to multiple agents, they performed tests with three copies of the same trained space-based observer, replicated into different orbits. Siew et al. [26] later extended this work to test similarly trained agents in scenarios featuring up to 800 near-GEO RSOs. However, given the current population of LEO, an operational-scale tasking and scheduling strategy needs to scale up to tens of thousands of RSOs. Greve [27] developed a multi-objective evolutionary algorithm tasker (MEAT) for tasking, but not scheduling, both ground-based and space-based SSA systems. Although this approach was successfully applied in simulations with tens of thousands of RSOs, as an evolutionary algorithm, MEAT does not provide any guarantees of optimality [27]. There is, however, no tasking and scheduling algorithm that is more scalable than the operational SP Tasker [23], and which provides some bounds or guarantees on performance as it relates to optimality.
To address these gaps, in recent work, we developed a novel tasking and scheduling formulation for the tracking of LEO RSOs using ground-based SSA sensors [28]. Based on the classical vehicle routing problem (VRP), this formulation leverages a set of sparse, binary feasibility matrices to represent the nonlinear dynamics and constraints of the catalog maintenance problem. We were able to demonstrate our formulation through simulations with 10,000 RSOs tracked by 27 sensors over 24 hours [28]. In parallel, we previously investigated the physics and other considerations that enable optical space-based SSA, with a focus on detecting RSOs that are challenging to reliably track using ground-based sensors [29,30].
In the present work, we synthesize our earlier contributions, applying models of optical physics and comparable tasking and scheduling algorithms to space-based tracking of RSOs that are larger than 10 cm in characteristic length. We focus on a case study involving taskable constellations of purpose-built, agile satellites, like Canada’s Sapphire spacecraft [2,9], with specialized, body-pointed optical sensors dedicated to the SSA mission. We introduce a modified sequential assignment problem (SAP) formulation for optimally tasking and scheduling these spacecraft to maximize the number of RSOs that can be tracked in a given time horizon. We demonstrate the capability of this formulation in operational-scale simulations featuring up to 20,000 RSOs, derived from real two-line element (TLE) sets, with up to 99.6% of targets successfully tracked by a constellation of 24 SB-SSA spacecraft over 24 hours. In smaller-scale simulations, we find that the dynamic spatial distribution of space-based systems enables a constellation of 6 such spacecraft to track 10% more LEO targets over a single orbital period than a network of 4 ground-based radar sensors over three orbital periods. We likewise find that a constellation of 12 such spacecraft can track several times as many LEO RSOs as a network of 12 ground-based optical sensors, which can be impacted by illumination, weather, and cloud cover. These results motivate investments in additional space-based assets for catalog maintenance, along with the tasking and scheduling algorithms needed to efficiently use them.
Date of Conference: September 16-19, 2025
Best Student Paper Award Winner 2025
Track: Space Domain Awareness