COA-GPT: Generative Pre-trained Transformers for Accelerated Course of Action Development in Military Operations

TL;DR

COA-GPT's capability to rapidly adapt and update COAs during missions presents a transformative potential for military planning, particularly in addressing planning discrepancies and capitalizing on emergent windows of opportunity.

Abstract

The development of Courses of Action (COAs) in military operations is traditionally a time-consuming and intricate process. Addressing this challenge, this study introduces COA-GPT, a novel algorithm employing Large Language Models (LLMs) for rapid and efficient generation of valid COAs. COA-GPT incorporates military doctrine and domain expertise to LLMs through in-context learning, allowing commanders to input mission information - in both text and image formats - and receive strategically aligned COAs for review and approval. Uniquely, COA-GPT not only accelerates COA development, producing initial COAs within seconds, but also facilitates real-time refinement based on commander feedback. This work evaluates COA-GPT in a military-relevant scenario within a militarized version of the StarCraft II game, comparing its performance against state-of-the-art reinforcement learning algorithms. Our results demonstrate COA-GPT's superiority in generating strategically sound COAs more swiftly, with added benefits of enhanced adaptability and alignment with commander intentions. COA-GPT's capability to rapidly adapt and update COAs during missions presents a transformative potential for military planning, particularly in addressing planning discrepancies and capitalizing on emergent windows of opportunities.

Figures (9)

• Figure 1: Overview of the proposed method for COA-GPT. COA-GPT consists of a large language model (LLM) that learns via in-context learning by initially being prompted with a knowledge base and additional constraints to be respected during COA development. Command and control personnel supply mission information, which is used by COA-GPT to generate options for COAs and iterates with humans via natural language until the final COA is selected.
• Figure 2: TF 1‒12 CAV Area of Operations in TigerClaw dsifirstyear.
• Figure 3: Satellite view (right) of the area of operations and its representation in StarCraft II (left) dsifirstyear.
• Figure 4: Example of image given to COA-GPT as part of mission information for experiments with LLMs with vision capabilities. The image overlays force arrangements in a satellite image of the battlefield terrain.
• Figure 5: Sample COA generated by COA-GPT for human review, including a visual representation, mission name, and strategy description.