Publicação
Learning to play DCSS with Deep Reinforcement Learning
| Resumo: | DCSS is a roguelike game in which the player must explore and find artifacts. In every step of the game, there are decisions to make, and the complexity of the game resides in the vast amount of options available to the player at any given time. The commands can be divided into classes such as movement, combat, inventory management and usage, spell casting, and divine abilities. We aim to implement an intelligent bot that will be able to play the game. To do so, we will use DRL. DRL is where deep learning and RL meets. It uses the same principles of RL, to learn to perform a task, receiving rewards for every action made. The difference is that the action we will perform is chosen by a DNN, and therefore, we call it DRL PyTorch[11] will be as the framework used to implement a NN that will be able to find the solution for every small decision that can be made during gameplay. If all of those decisions are merged, an intelligent bot should be able to play with some degree of success. The aim of the intelligent bot is to learn a task, which in this case, is playing the game, without programming any real behavior. |
|---|---|
| Autores principais: | Gonçalves, André Almeida |
| Assunto: | Dungeon Crawler Stone Soup Inteligência Artificial Aprendizagem por Reforço Deep Reinforcement Learning Artificial Inteligence Reinforcement Learning |
| Ano: | 2019 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | DCSS is a roguelike game in which the player must explore and find artifacts. In every step of the game, there are decisions to make, and the complexity of the game resides in the vast amount of options available to the player at any given time. The commands can be divided into classes such as movement, combat, inventory management and usage, spell casting, and divine abilities. We aim to implement an intelligent bot that will be able to play the game. To do so, we will use DRL. DRL is where deep learning and RL meets. It uses the same principles of RL, to learn to perform a task, receiving rewards for every action made. The difference is that the action we will perform is chosen by a DNN, and therefore, we call it DRL PyTorch[11] will be as the framework used to implement a NN that will be able to find the solution for every small decision that can be made during gameplay. If all of those decisions are merged, an intelligent bot should be able to play with some degree of success. The aim of the intelligent bot is to learn a task, which in this case, is playing the game, without programming any real behavior. |
|---|