Simulation to Strategy: Investigating P-LEO Satellite Internet Market Dynamics via Gaming and Reinforcement Learning

Abstract

An influx of private actors and state-owned agencies are entering the space industry to capitalize on space-based infrastructure and resources. Specifically, low Earth orbit (LEO) is seen as an untapped satellite communications (SATCOM) resource. With the deployment of SpaceX's Starlink, the concept of a proliferated low Earth orbit (P-LEO) satellite constellation for internet services has become a reality. These networks of thousands of satellites coordinate to deliver internet access directly to millions of user terminals. Despite mixed results among P-LEO operators, the joint market and orbital dynamics governing operational and orbital sustainability remain largely unexplored. This dissertation investigates these dynamics by modeling the P-LEO SATCOM market as a dynamical system to be explored.

First, using Satellite Tycoon, a tabletop board game developed as a multi-player simulation environment, a randomized controlled trial (RCT) is conducted to study how human participants develop constellation management strategies in response to economic and policy instruments. Results reveal that while players' revenue efficiency improves across repeated play, the tested policy treatments had a limited effect on satellite overproduction and debris generation, though derelict satellite debris showed a measurable reduction under the treatment condition. Next, a parameterized, single-agent reinforcement learning (RL) environment was constructed to model the specific dynamics and effects experienced by a single P-LEO constellation operator. An orbit plane catalog was created to give the agent a wide variety of constellation design choices, and RL agents were trained across a range of environmental configurations to explore which environment parameters most influence P-LEO constellation feasibility. Finally, the single-agent framework was extended into a multi-agent environment in which multiple RL agents act as constellation operators interacting and competing with one another for a limited market share. The utility function assigning customers to the single agent was replaced with a more sophisticated, reverse-bidding multi-attribute decision-making (MADM) model to more accurately model the zero-sum nature of customer acquisitions in SATCOM.