SpaceGym: Software Toolbox for Autonomous Systems in Noncooperative Space Operations

The SpaceGym software toolbox supports the development, training, and evaluation of autonomous space systems (e.g., satellites), with particular emphasis on noncooperative space operations. This toolbox is developed as part of the Multi-Agent Autonomous Space Technology (MAST) project within the Autonomous Systems (AS) Line portfolio. The SpaceGym toolbox covers the spectrum of autonomous space system technologies, including algorithms for autonomous perception, decision-making, and control; datasets and training environments for ML-based (machine learning) autonomous agents; and standardized, high-fidelity evaluation environments for benchmarking the performance of various autonomous agents.

The motivation for open-sourcing this toolbox comes from the disjoint, redundant, and nonreproduceable research efforts currently under development throughout the DoD and the space-autonomy community at large. A recent literature review and workshop on artificial intelligence (AI) for contested space revealed that, of the more than two dozen recently published works on noncooperative space operations, only one publication offered an open-source toolbox for evaluating autonomous space systems (and even that was very limited in scope) (Ravaioli, et al., 2022). This lack of common toolsets and evaluation benchmarks has led to a large amount of redundant work across the DoD where each enclave of researchers builds its own bespoke software tools to analyze common problems in noncooperative space operations. This redundancy and lack of common benchmarks has greatly hindered the advancement of autonomous and AI systems for space operations.

The SpaceGym toolbox seeks to remedy this situation by providing common toolsets, and ─ more importantly ─ common evaluation benchmark environments, for noncooperative space operations.