Year: 2025
Type: conference-paper
Abstract: In recent decades, thousands of satellites have been launched by governments and companies for scientific and commercial purposes, as the near-earth communication has advanced. However, the risk of collision with debris or other satellites is increasing as space becomes more crowded. Traditionally, collision avoidance is performed by operators on the ground who calculate debris conjunctions using mathematics methods. Recently, some automated methods have also been proposed by researchers. However, for satellite constellations, it is necessary to take coordination and cooperation into consideration. In this paper, we employ multi-agent reinforcement learning, an advanced artificial intelligent method, to automate the collision avoidance of satellites. We propose a multi-agent reinforcement learning method with communication constrained, which learns coordination and communication simultaneously. We compare our method and baselines on the proposed environment with different settings. The experiments prove that our method is efficient and prior to the baselines.
Authors: Yi Wang, Shengbo Chang, Hao Tang, Cheng Wang
Institution: Tongji University
Language: English
Subfield: Aerospace Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4413845184