Cooperative task scheduling and resource allocation of embodied multi-satellite systems: AI-driven perspective

In multi-satellite mission scheduling, an integrated space-ground architecture is typically employed, seamlessly integrating the brain-like ground-based offline planning (satellite-terrestrial collaboration) with the cerebellum-like onboard real-time coordination (multi-satellite collaboration). Within this framework, this survey comprehensively reviews cooperative task planning and resource allocation for embodied multi-satellite systems, leveraging cutting-edge artificial intelligence (AI), particularly reinforcement learning (RL) and large language models (LLMs), to overcome complex coordination challenges in strategic decision-making, synchronization across subsystems, and constrained resource optimization. First, a taxonomy of cooperative scheduling models, covering task scheduling and resource allocation, is presented for key mission profiles (disaster response, on-orbit servicing, deep-space exploration), highlighting pervasive spatial, temporal, and resource constraints. Then, AI-enhanced optimization algorithms in centralized, decentralized, and hybrid architectures are reviewed, classified as: (1) multi-objective offline optimization for mission deployment and strategic decisions, (2) online multi-objective optimization adapting to dynamic demands, and (3) explainable AI-driven scheduling ensuring trust in autonomous operations. Methodologies are validated via a high-fidelity simulation platform using digital twins and physics-informed neural networks for realistic emulation and benchmarking. Finally, future directions target fully autonomous AI agent swarms with collective reasoning, human-AI collaborative planning with enhanced interpretability, space autonomy, and multimodal LLM.