Dommele, Andre van (2024) Enhancing Football Simulation Performance in Deep Reinforcement Learning through Analytics-Based Dense Reward Shaping. Bachelor's Thesis, Artificial Intelligence.
|
Text
bAI2024vanDommeleA.pdf Download (1MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (155kB) |
Abstract
Recent research in reinforcement learning (RL) has seen a shift towards experimenting within complex and stochastic environments with large, dynamic state spaces. Football, one of the most popular sports, has garnered significant interest in RL research for these reasons. Football has also seen a rise in data-based analysis to improve real world player and team performance. A common issue in RL for complex environments is accurately modeling the relationship between the ultimate goal and the individual actions that produce desired behavior. Previous research shows that incorporating prior knowledge through reward shaping is essential for efficiently training RL agents to learn complex concepts, as it allows for conditioning individual actions at a local scale. Existing work either incorporates football domain knowledge within the model representation while keeping the sparse reward unchanged, or arbitrarily changes the reward without clear motivation. In this paper, we focus on incorporating football domain knowledge as a dense reward motivated by real world analytics in a proximal policy optimization-based RL scheme. Experimental results through extensive simulations against a fixed opponent show an improved policy in novel scenarios compared to the results published by the environment developers, as well as comparative performance to models with a greater architecture complexity.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Fernandes Cunha, R. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 13 Aug 2024 09:39 |
| Last Modified: | 13 Aug 2024 09:39 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/33947 |
Actions (login required)
 |
View Item |