Abstarct: In this thesis Reinforcement Learning (RL) have been used for investigating suitable policy in order to solve common problems in games such as inappropriate difficulty level and no versatility of games according to plaxyers’ capabilities. Learning Agent in games that uses Dyna Adaptive Learning percepts its environment by incorporating action-value function considering rewards. After that feature selection and organization, action and rewards (which exist in all games) have been discussed in combination with learning method. The learning agent start to learn against intelligent enemy agent (Rule-baxsed Algorithm). In the result section we proved that the learning agent have been shown better performance in comparison to existence methods for chosen policy in games. In this thesis two different tests have been discussed to approve purposed method. In the first one, the learning agent performance has been considered and the results show that by using the purposed algorithm the learning agent can finish the game successfully. In the second one, the purposed method have been compared with one of common methods in games.