Armed Services Problem
The US Armed Forces came to us with a sophisticated training tool that simulated real-world situations for design, experimentation, analysis, and training, but it was limited by human-designed rule-based behavior models. The predefined rules prevented the models from acting in optimal ways when a new situation emerged.
Solution
OpenTeams’ Open Source Architects introduced machine learning capability to the simulation by building a Python API and an MLOps pipeline that built reinforcement learning models. Reinforcement learning models exhibit autonomous behavior that allow simulation entities to adapt to complex situations, even if the human operator has not anticipated the situations. These models are exploratory and often find novel strategies that the human operator did not imagine.
Outcome
The new AI models create potential for new strategies, enhance training by offering a more interesting and engaging tool, provide new ways of experimenting and evaluating military technologies, and help plan for future needs.
Technologies Used
-
GeoPyGeoPy
-
PyTorchPyTorch
-
Stable-Baselines3Stable-Baselines3
-
PyMongoPyMongo
-
JupyterLabJupyterLab
