Deep reinforcement learning in RoboCup Keepaway

The central theme in multi-agent reinforcement learning often involves the coordination of agents to accomplish a common task. However, in complex environ- ments, the agent action space scales exponentially with the number of agents. As a result of this complexity, the coordination graph formalism allows the construction of a joint policy by decomposing the joint action values into an aggregation of local action-value functions. This approach is concerned with maximizing the joint opti- mal action of coordinating agents. In this thesis, we demonstrate and analyze how coordinated reinforcement learning can be used to learn a multi-agent policy on the RoboCup Keepaway domain, a sub-task of the RoboCup 2D soccer domain. This the- sis contributes to the enhanced implementation of the RoboCup Keepaway in Python and the Analysis of the Deep Coordination Graph and Deep Implicit Coordination Graph Algorithms.

Name thesis:degree_name

Master of Science in Computer Science

Discipline thesis:degree_discipline

Computer Science

Grantor

The University of Texas at Austin

Year dc:date.issued

2024

Author dc:creator

• Adekanmbi, Abayomi

Advisor dc:contributor.advisor

• Stone, Peter, 1971-

Committee member dc:contributor.committeemember

• Plaxton, C Greg

• Reinforcement learning

Language dc:language.iso

English

Identifier URI

https://doi.org/10.26153/tsw/61397

OAI identifier oai:identifier

oai:repositories.lib.utexas.edu:2152/134070