Abstract | ||
---|---|---|
We present an advanced version of dynamic scripting, which we apply to an agent created for the Fighting Game AI Competition. In contrast to the original method, our new approach is able to successfully adapt an agent's behavior in real-time scenarios. Based on a set of rules created with expert knowledge, a script containing a subset of these rules is created online to control our agent. Our method uses reinforcement learning to learn which rules to include in the script and how to arrange them. Results show that the algorithm successfully adapts the agent's behavior in tests against three other agents, allowing our agent to win most evaluations in our tests and the CIG 2014 competition. |
Year | DOI | Venue |
---|---|---|
2015 | 10.1007/978-3-319-24589-8_7 | Lecture Notes in Computer Science |
Keywords | Field | DocType |
Artificial Intelligence,AI,Computer Game,Fighting Game,Dynamic Scripting,Code Monkey,Real-Time,Adaptive,Reinforcement Learning | Computer science,Artificial intelligence,Multimedia,Reinforcement learning,Scripting language | Conference |
Volume | ISSN | Citations |
9353 | 0302-9743 | 5 |
PageRank | References | Authors |
0.45 | 11 | 3 |
Name | Order | Citations | PageRank |
---|---|---|---|
Kevin Majchrzak | 1 | 5 | 0.45 |
Jan Quadflieg | 2 | 43 | 6.52 |
Günter Rudolph | 3 | 219 | 48.59 |