SBIR Opportunity: Automated Course of Action Generation

Several approaches to CoA generation for military applications currently exist. These methods leverage large language models (LLM) applied to Doctrine (ARL’s COA-GPT, AI2C’s Camo-GPT, IBM’s Watson-X, Scale AI’s Donovan) or deep learning/hierarchical modeling applied to wargaming (ARL’s STRONG, Stilman Solutions’ LG-RAID, DARPA SCEPTER). Each existing LLM approach is inadequate for CoA generation at BN due to lack of proper consideration of Tactics, Techniques and Procedures (TTPs) and Standard Operating Procedures (SOPs) for both friendly and enemy forces. The existing learned/hierarchical approaches that depend on wargaming are also inadequate due to a lack of fully automated wargaming capability at the BN echelon and/or due to CoA outcomes that are not aligned with best practices (TTPs and SOPs). Proposals submitted under this topic will leverage state-of-the-art Artificial Intelligence (AI) approaches to create a CoA recommendation module narrowly tailored to the BN echelon. This capability will expand to encompass higher echelons in later phases.