Primitive auto-aim is used to create high quality initial training data for Q-learning, allowing agent to rapidly understand aiming.
Made some modifications to the raycasting map, converted the minimap to a heatmap that is refreshed and decays in real time.
I still need money to train this.
A fork of a simple raycasting example, I added a miniMap function that identifies cleared space, dead space, and walls and implements a decay function for cleared space to force agents to reprioritize it.