AI Agents Now Manage Fighter Formations in Simulation

RTDynamics demonstrated an advancement in its use of large language model (LLM)-based AI agents within the VR-Forces computer-generated forces (CGF) platform. In the demonstration, AI agents responded to natural language commands while collectively coordinating formations of fighter jets.

Each fighter jet in the simulation continues to be driven by its own AI agent, complete with an individual LLM connection and interaction history. What’s new is the ability for these agents to act collectively: coordinating maneuvers, sharing situational awareness, and executing formation-level tasks. This allows human operators to issue complex tactical instructions more naturally, without needing to repeat contextual information. Once a target or enemy is designated, for example, agents retain that knowledge and apply it to subsequent commands.

The demo scenario illustrates context persistence in the LLM-powered agents. When a formation is assigned a Combat Air Patrol (CAP) task—defining parameters such as waypoint (A), threat orientation (080∘ ), and altitude (30,000 ft)—these details are maintained in the shared context. Should the formation be diverted to engage hostile entities (bandits), an operator can later issue a simplified natural language command, such as "resume CAP," and the agents will automatically recall and execute the original configuration without the need to explicitly restate the waypoint, bearing, or altitude.

RTDynamics’ work built on VR-Forces’ use in military training, research, and operational analysis. The company’s simulation plugins include physics-based flight dynamics, air combat maneuvering, formation flight capabilities, and advanced modeling of sensors and weapons.

With formation-level coordination powered by LLM-based agents, RTDynamics is enabling next-generation experimentation in tactics, training, and AI-assisted decision-making.