Galapago

Using a simple model that is an understood and a relevant abstraction to the real-world scenario being considered, Galapago transforms modeling and simulation by making all game pieces intelligent and rapidly iterates through the decision space in a dynamic modeling and simulation environment. Galapago explains the range of good and bad outcomes as a result of the employed tactical and strategic decisions by the friendly agents and provides insights into the range of tactical and strategic decisions that the adversary agents may execute and the effect that those adversary decisions could have. Galapago is useful in gaining valuable insights that increase the decision makers' understanding of the friendly and adversary decision space and range of possible outcomes. Galapago can provide this insight to decision-makers before, during, and after real-world scenarios. Spear AI’s Galapago platform uses reinforcement learning as the basis of an explanatory model where agents, either manned or unmanned systems, operate in relevant peacetime, gray, or wartime environments, executing search, scouting, surveillance, and/or conflict scenarios. 

Galapago provides insight and understanding to decision makers for developing tactics, techniques, and procedures (TTPs), assessing acquisition platforms and systems, and operationally generating and providing multiple courses of action for both blue and red to consider. Galapago explains and provides insight into the range of good and bad outcomes as a result of learned tactical, operational, and strategic decisions by the friendly agents and adversary agents.

Our Reinforcement Learning Algorithm allow decision-makers to see everything from the random and uninformed to the optimal in a relevant environment and scenario. Galapago’s capabilities allow for the simulation of multi-domain operations (MDO). These currently include the air and sea domains and are scalable to other warfighting domains, such as land.
⠀
GalapagGo’s architecture uses a simple environment that can be understood and represents a relevant abstraction of the real-world scenario being considered. This “simple environment” allows for an understanding of which effects matter versus having effects confused because of model complexity, while maintaining a detailed enough level of abstraction that makes the environment relevant.
⠀
All units operating in the Galapago environment are intelligent, and the Galapago Reinforcement Learning algorithm rapidly iterates through the decision space for these units in the dynamic modeling and simulation environment. Galapago is currently able to run operational simulations thousands of times faster than real time.
Its underlying technology stack is based on open.ai’s framework and includes a user interface based on deck.gl, a training layer based on the PettingZoo and Ray libraries, an analytics layer leveraging WanDB, and scalable cloud compute with AWS SageMaker. GalapaGo is useful in gaining valuable insights that increase the decision makers' understanding of the friendly and adversary decision space and range of possible outcomes. Galapago can provide this insight to decision-makers before, during, and after real-world scenarios. Galapago is not a prediction algorithm but can be used to help forecast probable outcomes given other inputs such as historical data.