Guanchu He, University of Colorado Colorado Springs; Philip Brown, University of Colorado Colorado Springs; Kaylee Barcroft, Astra Ultra; Jeffrey Smith, Astra Ultra; Matthew Bonn, Astra Ultra; Josue Cardoso dos Santos, University of Colorado Colorado Springs; Jackson Thorne, University of Colorado Colorado Springs
Keywords: Co-orbital ASAT, game theory, sda, rpo
Abstract:
Space Domain Awareness (SDA) requires real-time assessment of satellite maneuvers to distinguish benign trajectory adjustments from potentially hostile actions. Currently, intent classification relies on manual analysis, which is slow, labor-intensive, and susceptible to ambiguity. Furthermore, if an intelligent adversary knows that it is being monitored, it can potentially use complicated sequences of maneuvers to attempt to mask hostile maneuvers as benign. Alternatively, an adversary could deceptively attempt to mask benign maneuvers as hostile in an attempt to desensitize a defender’s intent classifier. Either of these approaches may substantially add to the defender’s intent classification challenge. To address this challenge, we propose a game-theoretic framework that models maneuver interactions between a defending satellite (“Blue”) and a potentially adversarial satellite (“Red”) which may intend to execute a kinetic attack against Blue. This framework will enable automated inference of maneuver intent by formulating the interaction as a game with asymmetric information.
In this model, Red transitions from a benign state to a threatening state at an unknown time, while Blue must infer this transition by observing Red’s kinematics. We propose several baseline parameterized alert strategies for Blue which react to Red’s motion history and possible future trajectories, including a range of approaches from simple threshold-based triggers to advanced receding-horizon methods that estimate Red’s minimum time-to-collision. In response, Red can probe these alert strategies and attempt to induce false positive alerts early in the interaction or execute a stealthy attack late in the interaction. We adopt a leader-follower (“Stackelberg”) style interaction, in which Blue pre-commits to an alert strategy and then Red responds to attempt to maximize the probability of a successful kinetic attack at the end of the interaction. We constrain the actions of both satellites (with hard delta-v and rate limits on Red’s maneuvers and limits on the total time over which Blue can alert). We believe that this game-theoretic modeling approach offers a principled approach to maneuver intent classification.
To begin to instantiate the model in a way that uses real-world data, we propose a novel intent classification method which ingests TLE updates for a Red/Blue satellite pair and uses solutions to Lambert’s problem to identify future opportunities for the Red satellite to execute a kinetic attack on the Blue satellite. We then validate our approach by applying it to several historical examples of known RPO events. For instance, the 2009 HTV-1 resupply mission to the International Space Station has a list of known maneuvers which we can use as a ground truth to assess the effectiveness of our intent classifier. We show that our techniques correctly classify all of the known maneuvers made by HTV-1 during its approach to the ISS.
By leveraging this framework, we demonstrate how real-time data assimilation techniques—coupled with strategic decision models—can enhance SDA automation. Our results suggest that incorporating adversarial reasoning into SDA algorithms can significantly improve automated threat assessment, reducing reliance on manual intervention while maintaining operational accuracy. This research has direct implications for space security, as it provides a principled method for autonomous human-on-the-loop decision support in SDA. Future work will focus on integrating this framework with advanced strategic modeling and refining alert thresholds through reinforcement learning techniques. In addition, the modularity of our approach will enable us to study a wide variety of attack models.
Date of Conference: September 16-19, 2025
Track: Space Domain Awareness