Reinforcment Learning for Gameplaying
Time: Mon 2021-09-27 10.00
Participating: Amir H. Payberah
Reinforcement Learning (RL) is becoming ever more prevalent in game development.
However, there exist many challenges to overcome in order to use RL to its full potential.
For example, an RL agent trained in one game environment cannot easily be generalized
to replicate the same level of performance in new game environments with different
levels and variations. Moreover, the sparsity of extrinsic rewards of games, where
agents get rewards from the environments only at the end of game levels, makes it more
challenging to use RL agents in gameplaying. Several studies have shown that using
intrinsic rewards, which do not necessarily have the same objective as the game level, is
an effective way to deal with such cases. Intrinsic rewards allow for incorporating basic
skills in agent policies to better generalize over various game levels. In this lecture, I will
discuss the challenges of selecting basic skills with intrinsic rewards for match-3 games,
such as Candy Crush Friends Saga. Moreover, I will discusses how by combining basic
skills, we can improve the performance of RL models.