MLAHelper is the shared result of me and my colleague Toto Christensen, as our Master Thesis. It is a suite made for Unity, particularly for project that make use of Unity’s ML-Agents package.
Due to the package’s lack of transparency regarding the inner structure of the agent’s model, the tool presents a clear visualization of such model overlayed on the same screen where the game is shown. Additionally, the suite provides its users with components and systems that speed up the overall setup process of ML-Agents in a project.
The main features are the visualization, the observation manager, and the action-mask manager.