The player would then be able to select from the top few decisions ranked in order of success given the match statistics. If the health is below a certain threshold then the AI can be set to run away from the player and avoid it until another function is triggered. For example, players may be over or under performing in this game or may simply have a preference for the way they play often defined by the types of characters they play.
The AI is able to cheat at the difficulty, cheater 1 visionwhere it begins to have vision assistance on your location and units. With this feature, the player can actually consider how to approach or avoid an enemy.
First, I will list the main improvements that need to be made before this could be viable for implementation: Calculate the MDP using more data that represents the whole player population, not just competitive matches. We have introduced rewards for influencing the model output but how is this obtained?
However, there is still a need for improvement in this area.
Specifically, AI agents that are programmed to respond to hostile attacks will sometimes attack each other if their cohort's attacks land too close to them.