PhD Thesis Proposal 

Ricardo Silva

(Faculty Advisor: Professor Dimitri Mavris)

"A Multi-Agent Autonomous Framework For Satellite Traffic Management In Mega-Constellation"

Wednesday, June 3

11:00 a.m.

Weber, CoVE

Abstract: 

The rapid growth of satellite mega-constellations in low Earth orbit (LEO), including systems like Starlink, created substantial challenges for sustainable satellite traffic management (STM). As constellations expand from dozens to tens of thousands of satellites, the rate of conjunction events and collision avoidance maneuvers (CAM) rises significantly, beyond the scalability constraints of conventional human-centered operating methods. Current Conjunction Assessment Risk Analysis (CARA) methodologies mostly depend on static two-dimensional Probability of Collision (PoC) thresholds, whereas existing deterministic maneuver planning strategies fail to consider secondary collision risks and payload operational impact. These constraints drive the development of autonomous, scalable, and context-sensitive STM frameworks designed to support future mega-constellation operations.

This research introduces a novel autonomous STM framework consisting of two connected research domains: collision risk evaluation and autonomous collision avoidance maneuver planning. The initial study domain presents a hybrid CARA framework that combines three-dimensional PoC computations with Fuzzy Inference Systems (FIS) and Evidence-Based Reasoning (EBR). The proposed approach dynamically assesses the validity of conventional 2D PoC assumptions and integrates multiple evidence sources, including conjunction geometry and object-specific data from the available databases, into unified and interpretable collision risk assessment. Replacing single-threshold decision models with a multi-evidence reasoning method enhances situational awareness and minimizes wasteful moves.

The second study topic centers on autonomous Collision Avoidance Maneuver (CAM) planning via Multi-Agent Reinforcement Learning (MARL). A Multi-Agent Proximal Policy Optimization (MAPPO) framework is proposed where satellites act as autonomous agents, capable in autonomously apply collision avoidance techniques. The framework concurrently addresses secondary conjunction issue that result from CAM and degradation of ISL network performance due to orbital changes. To do this, metrics including propellant consumption, miss distance, secondary collision probability, and ISL availability are integrated into the agents' rewards frameworks and state representations. Pareto-based multi-objective optimization allows the framework to balance collision safety and communication network integrity across large constellations.

The methodology uses agent-based modeling with real conjunction data from Starlink constellation proximity events and Starlink reports from the first quarter of 2026. The proposed frameworks are evaluated against existing operating procedures using metrics such as PoC reduction, delta-V consumption, secondary encounter rate, and ISL availability. This study improves current capabilities in autonomous satellite traffic management and enables sustainable operations of mega-constellations.

Committee:
Dr. Dimitri Mavris (advisor), School of Aerospace Engineering
Dr. Karen Feigh, School of Aerospace Engineering
Dr. Thomas Roberts, School of Aerospace Engineering, School of International Affairs
Dr. Bradford Robertson , School of Aerospace Engineering
Dr. Stephanie Introne, The Aerospace Corporation