AI-Driven Autonomous Robotics for Scalable Lunar and Martian Infrastructure Optimizing ISRU Swarm Construction and Decentralized Control
- Paper ID
101540
- DOI
- author
- company
West Virginia University; Space Generation Advisory Council (SGAC); ; Drexel University
- country
United States
- year
2025
- abstract
Achieving sustained extraterrestrial habitation requires shifting from Earth-dependent logistics to AI-driven, autonomous robotic systems capable of in-situ resource extraction, on-site manufacturing, and adaptive infrastructure deployment. This work presents a robotic framework integrating machine learning-optimized autonomy, ISRU-based material processing, and robotic assembly to establish scalable, self-sustaining infrastructure on the Moon, Mars, and beyond. The proposed architecture minimizes payload constraints, optimizes resource utilization, and enhances mission resilience through autonomous decision-making, adaptive robotic coordination, and decentralized infrastructure management. The system comprises three key subsystems: (1) AI-driven robotic excavation and processing platforms for ISRU-based material extraction, (2) multi-agent robotic assembly teams for constructing planetary surface infrastructure, and (3) a decentralized AI-based control architecture for dynamic mission adaptability. The ISRU subsystem integrates imaging fusion and subsurface tomography-based material mapping, autonomously detecting regolith deposits and water ice reservoirs. A reinforcement learning-based excavation policy enables robotic platforms to optimize excavation force, energy consumption, and tool trajectories based on soil composition and mechanical resistance. For infrastructure assembly, swarm-based robotic construction systems leverage multi-degree-of-freedom manipulators, incorporating force-adaptive feedback and imitation learning to ensure precision in modular component placement, interlocking, and adaptive structural augmentation. A multi-agent task allocation algorithm, utilizing deep multi-agent reinforcement learning (MARL), optimizes dynamic task execution and collaborative behaviors in planetary environments. The modular lattice-based construction paradigm facilitates self-repairing, autonomously expandable structures, ensuring long-term mission sustainability. A multi-layered adaptive communication and navigation network, leveraging federated learning for mesh-network optimization and quantum-resilient cryptographic communication, ensures high-reliability data exchange between surface assets, orbital stations, and Earth-based command centers. Validation of the architecture is conducted through physics-based simulations incorporating NASA’s LROC and HiRISE-derived lunar and Martian terrain datasets. AI-driven robotic excavation, material processing efficiency, and construction precision are benchmarked against real-world regolith excavation datasets and validated in laboratory-scale experiments using lunar regolith analogs. Preliminary findings indicate that ISRU-based infrastructure deployment can reduce payload mass by 28%, compared to conventional pre-fabrication methods, while AI-driven robotic assembly improves deployment efficiency by 38% due to enhanced multi-agent coordination, real-time adaptive path planning, and autonomous fault recovery mechanisms. Future work will focus on advancing deep reinforcement learning models for real-time robotic perception under variable gravity conditions, developing self-healing composite materials for infrastructure resilience, and enhancing human-AI cooperative frameworks for autonomous planetary surface operations. Keywords Reinforcement Learning, Multi-Agent Systems, Modular Habitat Construction, Planetary Infrastructure, Adaptive Mesh Networking, Decentralized Control Architectures