Levita, Luca
(2023)
DefaultVR: the AI Expansion. An application of artificial intelligence in competitive gaming and virtual reality.
[Laurea magistrale], Università di Bologna, Corso di Studio in
Artificial intelligence [LM-DM270]
Documenti full-text disponibili:
Abstract
This works presents the development of DefaultVR: The AI Expansion, an
expansion of the first degree thesis DefaultVR, a virtual reality tactical shooter
online game. In particular, this expansion aims to include an artificial agent,
called Eve, capable of learning to play in the virtual reality world through
reinforcement learning techniques. The agent learns to navigate the game
environment, make decisions based on pseudo-visual information, and optimize its actions to maximize rewards. The development utilizes a deep
reinforcement learning framework with the Proximal Policy Optimization
algorithm included with Units’s ML-Agents.
Extensive experiments were conducted to evaluate the agent’s performance, comparing it against itself and human players. The results demonstrate the agent’s ability to adapt and improve over time, achieving competitive gameplay skills comparable to both new and experienced human
VR players. The training process involved iterative optimization and analysis of various hyperparameters, observations’ and actions’ spaces, and
training configurations.
The successful development of the artificial agent has significant implications for the field of gaming AI, showcasing its potential for creating engaging and challenging gameplay experiences. The research contributes to
the broader understanding of reinforcement learning techniques and their
application in training intelligent agents for real-world tasks.
Abstract
This works presents the development of DefaultVR: The AI Expansion, an
expansion of the first degree thesis DefaultVR, a virtual reality tactical shooter
online game. In particular, this expansion aims to include an artificial agent,
called Eve, capable of learning to play in the virtual reality world through
reinforcement learning techniques. The agent learns to navigate the game
environment, make decisions based on pseudo-visual information, and optimize its actions to maximize rewards. The development utilizes a deep
reinforcement learning framework with the Proximal Policy Optimization
algorithm included with Units’s ML-Agents.
Extensive experiments were conducted to evaluate the agent’s performance, comparing it against itself and human players. The results demonstrate the agent’s ability to adapt and improve over time, achieving competitive gameplay skills comparable to both new and experienced human
VR players. The training process involved iterative optimization and analysis of various hyperparameters, observations’ and actions’ spaces, and
training configurations.
The successful development of the artificial agent has significant implications for the field of gaming AI, showcasing its potential for creating engaging and challenging gameplay experiences. The research contributes to
the broader understanding of reinforcement learning techniques and their
application in training intelligent agents for real-world tasks.
Tipologia del documento
Tesi di laurea
(Laurea magistrale)
Autore della tesi
Levita, Luca
Relatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
Deep Reinforcemente Learning,Unity,ML-Agents,PPO,MA-POCA,Virtual Reality,Physics Simulation
Data di discussione della Tesi
20 Luglio 2023
URI
Altri metadati
Tipologia del documento
Tesi di laurea
(NON SPECIFICATO)
Autore della tesi
Levita, Luca
Relatore della tesi
Scuola
Corso di studio
Ordinamento Cds
DM270
Parole chiave
Deep Reinforcemente Learning,Unity,ML-Agents,PPO,MA-POCA,Virtual Reality,Physics Simulation
Data di discussione della Tesi
20 Luglio 2023
URI
Statistica sui download
Gestione del documento: