• The Journal of Korea Robotics Society
• /
• v.14 no.3
• /
• pp.157-162
• /
• 2019

The video game Tetris is one of most popular game and it is well known that its game rule can be modelled as MDP (Markov Decision Process). This paper presents a DQN (Deep Q-Network) based game agent for Tetris game. To this end, the state is defined as the captured image of the Tetris game board and the reward is designed as a function of cleared lines by the game agent. The action is defined as left, right, rotate, drop, and their finite number of combinations. In addition to this, PER (Prioritized Experience Replay) is employed in order to enhance learning performance. To train the network more than 500000 episodes are used. The game agent employs the trained network to make a decision. The performance of the developed algorithm is validated via not only simulation but also real Tetris robot agent which is made of a camera, two Arduinos, 4 servo motors, and artificial fingers by 3D printing.

• Journal of Digital Convergence
• /
• v.14 no.8
• /
• pp.293-303
• /
• 2016

Analysis of bargaining games utilizing evolutionary computation in recent years has dealt with important issues in the field of game theory. In this paper, we investigated interaction and coevolution process among heterogeneous artificial agents using evolutionary computation in the bargaining game. We present three kinds of evolving-strategic agents participating in the bargaining games; genetic algorithms (GA), particle swarm optimization (PSO) and differential evolution (DE). The co-evolutionary processes among three kinds of artificial agents which are GA-agent, PSO-agent, and DE-agent are tested to observe which EC-agent shows the best performance in the bargaining game. The simulation results show that a PSO-agent is better than a GA-agent and a DE-agent, and that a GA-agent is better than a DE-agent with respect to co-evolution in bargaining game. In order to understand why a PSO-agent is the best among three kinds of artificial agents in the bargaining game, we observed the strategies of artificial agents after completion of game. The results indicated that the PSO-agent evolves in direction of the strategy to gain as much as possible at the risk of gaining no property upon failure of the transaction, while the GA-agent and the DE-agent evolve in direction of the strategy to accomplish the transaction regardless of the quantity.

• Journal of Information Processing Systems
• /
• v.16 no.5
• /
• pp.1074-1082
• /
• 2020

Recently, research has been actively conducted to create artificial intelligence agents that learn games through reinforcement learning. There are several factors that determine performance when the agent learns a game, but using any of the activation functions is also an important factor. This paper compares and evaluates which activation function gets the best results if the agent learns the game through reinforcement learning in the 2D racing game environment. We built the agent using a reinforcement learning algorithm and a neural network. We evaluated the activation functions in the network by switching them together. We measured the reward, the output of the advantage function, and the output of the loss function while training and testing. As a result of performance evaluation, we found out the best activation function for the agent to learn the game. The difference between the best and the worst was 35.4%.

• Journal of Korea Game Society
• /
• v.19 no.6
• /
• pp.61-70
• /
• 2019

In this paper, we apply reinforcement learning in the field of management simulation game to check whether game agents achieve autonomously given goal. In this system, we apply PPO (Proximal Policy Optimization) algorithm in the Unity Machine Learning (ML) Agent environment and the game agent is designed to automatically find a way to play. Five game scenario simulation experiments were conducted to verify their usefulness. As a result, it was confirmed that the game agent achieves the goal through learning despite the change of environment variables in the game.

• Journal of Korea Game Society
• /
• v.9 no.5
• /
• pp.117-125
• /
• 2009

In real-time strategy simulation (RTS) game each team is composed of agents and executes strategies to win other team. Strategy needs agents' cooperation in each team. This needs multi-agent system (MAS). Centralized decision making, one of decision making in MAS, selects actions not to agents but to team by a coordinated agent. Decentralized decision making costs high because each agent communicates with each other. In this paper we propose a system which controls agents by grouping and allocates roles through negotiation by a coordinated agent. Then, when one of allocated actions is not executed or failed, a coordinated agent allocates its role to another agent. We make experiments in starcraft, famous RTS game. When a proposed method is applied, the performance of attack and defense is increased. The improved agents' team wins eight times per ten games.

• Smart Media Journal
• /
• v.13 no.2
• /
• pp.32-43
• /
• 2024

Game agents which are behavioral agent for game playing are a crucial component of game satisfaction. However it takes a lot of time and effort to create game agents for various game levels, environments, and players. In addition, when the game environment changes such as adding contents or updating characters, new game agents need to be developed and the development difficulty gradually increases. And it is important to have a game agent that can be customized for different levels of players. This is because a game agent that can play games of various levels is more useful and can increase the satisfaction of more players than a high-level game agent. In this paper, we propose a method for learning and controlling the level of play of game agents that can be rapidly developed and fine-tuned for various game environments and changes. At this time, reinforcement learning applies a policy-based distributed reinforcement learning method IMPALA for flexible processing and fast learning of various behavioral structures. Once reinforcement learning is complete, we choose actions by sampling based on Softmax-Temperature method. From this result, we show that the game agent's play level decreases as the Temperature value increases. This shows that it is possible to easily control the play level.

• Information Systems Review
• /
• v.23 no.1
• /
• pp.1-22
• /
• 2021

This study aims to analyze online game user's game items transacting behaviors for the two game genres such as MMORPG and sports game. For the sake of conducting the analysis, we adopted a fuzzy logic agent-based modeling. In the online game fields, game items transactions are crucial to game company's profitability. However, there are lack of previous studies investigating the online game user's game items transacting activities. Since many factors need to be addressed in a complicated way, ABM (agent-based modeling) simulation mechanism is adopted. Besides, a fuzzy logic is also considered due to the fact that a number of uncertainties and ambiguities exist with respect to online game user's complex behaviors in transacting game items. Simulation results from applying the fuzzy logic ABM method revealed that MMORPG game users are motivated to pay expensive price for high-performance game items, while sports game users tend to transact game items within a reasonable price range. We could conclude that the proposed fuzzy logic ABM simulation mechanism proved to be very useful in organizing an effective strategy for online game items management and customers retention.

• Journal of Convergence for Information Technology
• /
• v.9 no.2
• /
• pp.108-114
• /
• 2019

In this paper, we propose an intelligent vocabulary recommendation agent that automatically provides vocabulary corresponding to game-based learners' needs and requirements in the mobile education augmented reality game environment. The proposed agent reflects the characteristics of mobile technology and augmented reality technology as much as possible. In addition, this agent includes a vocabulary reasoning module, a single game vocabulary recommendation module, a battle game vocabulary recommendation module, a learning vocabulary list Module, and a thesaurus module. As a result, game-based learners' are generally satisfied. The precision of context vocabulary reasoning and thesaurus is 4.01 and 4.11, respectively, which shows that vocabulary related to situation of game-based learner is extracted. However, In the case of satisfaction, battle game vocabulary(3.86) is relatively low compared to single game vocabulary(3.94) because it recommends vocabulary that can be used jointly among recommendation vocabulary of individual learners.

• Journal of Korea Game Society
• /
• v.7 no.4
• /
• pp.93-104
• /
• 2007

In the complex real-time multi-agent system such as game environment, dynamic task allocations are repeatedly performed to achieve a goal in terms of system efficiency. In this research, we present a task allocation scheme suitable for the real-time multi-agent environment. The scheme is to optimize the task allocation by complementing existing coordination agent with $A^*$ algorithm. The coordination agent creates a status graph that consists of nodes which represent the combinations of tasks and agents, and refines the graph to remove nodes of non-execution tasks and agents. The coordination agent performs the selective utilization of the $A^*$ algorithm method and the greedy method for real-time re-allocation. Then it finds some paths of the minimum cost as optimized results by using $A^*$ algorithm. Our experiments show that the coordination agent with $A^*$ algorithm improves a task allocation efficiency about 25% highly than the coordination agent only with greedy algorithm.

• Journal of Korea Game Society
• /
• v.19 no.6
• /
• pp.5-14
• /
• 2019

In this paper, we propose a modeling and simulation of group behavioral movement of pedestrians using Agent based and Cellular Automata model in intelligent game environment. The social behaviors of the crowds are complex and important, and based on this, the prototype game-model was implemented to show the crowd interaction on AB and CA in the game environment. Our experiment revealed the promise of group behaviour as a cost-efficient, yet accurate platform for researching crowd behaviour in risk situations with real models.