we designed a reward function that isbased on a game-balancing constant and introduce itinto the Proximal-Policy-Opmitization (PPO) (Schul-man et al., 2017) algorithm, a reinforcement learn-ing method that directly optimizes the policy usinggradient-based learning.

still because the player may have a wrong vision ofits own abilities (Missura and G ̈artner, 2009)

remains inside a range around this constant during the training

a reward function based on a balancingconstan

how to act while still maintaining the balancing