Army Cyber unveils AGAMOTTO tactical AI
The Army Cyber Institute at West Point introduced AGAMOTTO, an AI framework pairing reinforcement learning with Unity-based simulation to generate tactical courses of action from Tactical Assault Kit data. Initial benchmarks on local consumer hardware confirm real-time simulation is feasible, but reveal pure reinforcement learning cannot produce viable tactical maneuvers within compressed operational timelines.
While strategic defense AI focuses on cloud-scale LLMs, AGAMOTTO proves that tactical-edge autonomy faces steep algorithmic bottlenecks under strict hardware and time constraints.
- –Cursor on Target (CoT) telemetry from ATAK ingests directly into Unity via RIDE, automating scenario reconstruction without manual map entry.
- –Benchmarking on a single NVIDIA RTX 5090 desktop reflects realistic forward-deployed compute rather than reliance on unreachable cloud clusters.
- –Soft Actor-Critic (SAC) models failed to discover survivable attack routes within a 6-hour tactical planning window, demonstrating the limits of unguided exploration.
- –Effective course-of-action generation will require hybrid imitation learning and expert heuristics to prune policy search spaces before AI can guide live tactical maneuvers.
DISCOVERED
3h ago
2026-11-20
PUBLISHED
3h ago
2026-11-20
RELEVANCE
AUTHOR
Discover AI