Artificial Intelligence in Combat Decision-Making: Weapon Target Assignment via Reinforcement Learning and Graph...

Area of Science:

• Artificial Intelligence
• Operations Research
• Military Science

Background:

• Dynamic weapon-target assignment (DWTA) is critical for battlefield success.
• Current deep reinforcement learning (DRL) methods for DWTA face challenges in representing battlefield topology, scalability, and performance metric relevance.

Purpose of the Study:

• To address limitations in existing DWTA approaches.
• To develop an improved DWTA solution using DRL and graph neural networks (GNNs).

Main Methods:

• A novel partially observable Markov decision process (POMDP) framework was designed.
• Incorporated graph-based action representation, observation features, and reward design.
• Leveraged DRL and GNNs for advanced decision-making.

Main Results:

• The proposed DRL and GNN approach demonstrated effectiveness across naval and ground combat domains.
• Outperformed existing heuristic and meta-heuristic methodologies in DWTA.
• Comprehensive validation confirmed the efficacy of the GNN and decision-making patterns.

Conclusions:

• The integrated DRL and GNN framework offers a significant advancement in solving the DWTA problem.
• The novel POMDP design enhances scalability and relevance for complex military operations.
• This approach provides a more robust and effective solution for strategic battlefield decision-making.