An Action Select Reinforcement Learning Model in Multi Agent Environment
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
Reinforcement learning is concerned with how an agent ought to take actions in an environment so as to maximize some notion of long-term reward. In particular, a multi-agent system consisting of multiple interacting agents has increased state and action space as compared to a single agent system, so it must have an effective action selection mechanism. Reinforcement learning is used to evaluate and improve the effectiveness of a robotic soccer agent's action selection. That is, an agent that chooses its actions according to a certain action selection policy receives feedback regarding whether the chosen actions are desirable or not, and the agent learns to find optimal actions for various states in simulated soccer games based on the feedback. Possible states were identified by analyzing various situations/conditions arising in simulated soccer games, and then action selection policies were defined based on the analysis of a soccer agent's behavior patterns. In this paper can be exploited to develop optimized strategies and tactics for robot soccer systems, and it is also applicable to other multi-agent learning environments similar to the robot soccer game environment. In such environments, it is significant to acquire a policy that enables intelligent agents to work cooperatively to win the game.
Abstract
Reinforcement learning is concerned with how an agent ought to take actions in an environment so as to maximize some notion of long-term reward. In particular, a multi-agent system consisting of multiple interacting agents has increased state and action space as compared to a single agent system, so it must have an effective action selection mechanism. Reinforcement learning is used to evaluate and improve the effectiveness of a robotic soccer agent's action selection. That is, an agent that chooses its actions according to a certain action selection policy receives feedback regarding whether the chosen actions are desirable or not, and the agent learns to find optimal actions for various states in simulated soccer games based on the feedback. Possible states were identified by analyzing various situations/conditions arising in simulated soccer games, and then action selection policies were defined based on the analysis of a soccer agent's behavior patterns. In this paper can be exploited to develop optimized strategies and tactics for robot soccer systems, and it is also applicable to other multi-agent learning environments similar to the robot soccer game environment. In such environments, it is significant to acquire a policy that enables intelligent agents to work cooperatively to win the game.
참고문헌 (12)
- Matthew E. Taylor, "Representation transfer for reinforcement learning", AAAI 2007 Fall Symposium (2007)
- Samuel Barrett, "Cooperating with unknown teammates in complex domains:A robot soccer case study of Ad hoc teamwork", In Proceedings of the Twenty-Ninth AAAI Conference on Artificial Intelligence (2015)
- Multi-agent reinforcement learning for multi-robot system: A survey, "Yang, E.", University of Essex, ~, pp.CSM-404 (2004)
- Piyush Khandelwal, "Multi-robot human guidance using topological graphs", In AAAI Spring 2014 Symposium on Qualitative Representations for Robots (AAAI-SSS) (2014)
- Alexander A. Sherstov, "Function approximation via tile coding:Automating parameter choice", In Proc. Symposium on Abstraction, Reinforcement, and Approximation(SARA-05) (2005)
- Noa Agmon, "Leading Ad hoc agents in joint action settings with multiple teammates", In Proc. of 11th Int. Conf. on Autonomous Agents and Multiagent Systems (AAMAS) (2012)
- Sutton, R.S., "Reinforcement learning: An introduction" (1998), MIT Press
- Fangkai Yang, "Planning in answer set programming while learning action costs for mobile robots", In AAAI Spring 2014Symposium on Knowledge Representation and Reasoning in Robotics (AAAI-SSS) (2014)
- Whiteson, S., "Concurrent layered learning", pp.193~200 (2003), ACM Press
- David McAllester, "Keeping the ball from CMUnited-99, Robocup-2000:Robocup IV" (2001), Springer Verlag
- "The CMA evolution strategy: A Tutorial" (2009)[URL]
- Patrick MacAlpine, "The RoboCup 2013drop-in player challenges: Experiments in Ad hoc teamwork", In Proceedings of the IEEE, 2014, ~