Towards Realistic Earth-Observation Constellation Scheduling: Benchmark and Methodology

TL;DR

This work tackles realistic scheduling for Agile Earth Observation Satellite constellations by introducing AEOS-Bench, a large-scale, physics-enabled benchmark with ground-truth annotations, and AEOS-Former, a constraint-aware Transformer scheduler. The model uses a dedicated internal constraint module and a simulation-driven iterative learning loop to respect physical limits while optimizing task completion and energy efficiency. Across multiple splits, AEOS-Former outperforms baselines and ablations confirm the contribution of its constraint-guided attention and iterative training. By releasing both the benchmark and the model, the authors enable reproducible evaluation and foster further advances in AEOS constellation scheduling.

Abstract

Agile Earth Observation Satellites (AEOSs) constellations offer unprecedented flexibility for monitoring the Earth's surface, but their scheduling remains challenging under large-scale scenarios, dynamic environments, and stringent constraints. Existing methods often simplify these complexities, limiting their real-world performance. We address this gap with a unified framework integrating a standardized benchmark suite and a novel scheduling model. Our benchmark suite, AEOS-Bench, contains $3,907$ finely tuned satellite assets and $16,410$ scenarios. Each scenario features $1$ to $50$ satellites and $50$ to $300$ imaging tasks. These scenarios are generated via a high-fidelity simulation platform, ensuring realistic satellite behavior such as orbital dynamics and resource constraints. Ground truth scheduling annotations are provided for each scenario. To our knowledge, AEOS-Bench is the first large-scale benchmark suite tailored for realistic constellation scheduling. Building upon this benchmark, we introduce AEOS-Former, a Transformer-based scheduling model that incorporates a constraint-aware attention mechanism. A dedicated internal constraint module explicitly models the physical and operational limits of each satellite. Through simulation-based iterative learning, AEOS-Former adapts to diverse scenarios, offering a robust solution for AEOS constellation scheduling. Experimental results demonstrate that AEOS-Former outperforms baseline models in task completion and energy efficiency, with ablation studies highlighting the contribution of each component. Code and data are provided in https://github.com/buaa-colalab/AEOSBench.

Figures (9)

• Figure 1: Illustration of AEOS constellation scheduling over four timesteps. At each timestep, satellites adjust their attitude to image ground targets, consuming battery energy while charging via solar panels. Tasks can be published or expired. Multiple satellites can cooperate to complete tasks.
• Figure 2: The generation process of satellite assets, which incorporates an empirical formula and multiple checks to ensure stable attitude control for each asset.
• Figure 3: The annotation pipeline for AEOS-Bench.
• Figure 4: Statistical analysis of AEOS-Bench. (a) and (b) show the distribution of trajectories w.r.t. the semi-major axis and eccentricity of satellite orbits, respectively. (c) and (d) illustrate the distribution of trajectories w.r.t. the number of satellites and tasks, respectively.
• Figure 5: Architecture of AEOS-Former. Static and dynamic data of satellites and tasks are first concatenated and embedded. A transformer encoder processes task features, and a decoder attends to satellite embeddings under a constraint-derived cross‐attention mask. The internal constraint module then predicts feasibility logits and required control times, guiding action selection.