Cooperation without communication
Distributed Artificial Intelligence
RoboCup: The Robot World Cup Initiative
AGENTS '97 Proceedings of the first international conference on Autonomous agents
Team-partitioned, opaque-transition reinforcement learning
Proceedings of the third annual conference on Autonomous Agents
Introduction to Reinforcement Learning
Introduction to Reinforcement Learning
Reinforcement Learning in the Multi-Robot Domain
Autonomous Robots
Scaling Reinforcement Learning toward RoboCup Soccer
ICML '01 Proceedings of the Eighteenth International Conference on Machine Learning
Multiagent Reinforcement Learning: Theoretical Framework and an Algorithm
ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
Learning in Multi-Robot Systems
IJCAI '95 Proceedings of the Workshop on Adaption and Learning in Multi-Agent Systems
Karlsruhe Brainstormers - A Reinforcement Learning Approach to Robotic Soccer
RoboCup 2000: Robot Soccer World Cup IV
Practical reinforcement learning in continuous domains
Practical reinforcement learning in continuous domains
Layered learning in multiagent systems
Layered learning in multiagent systems
Hi-index | 0.00 |
With bottom-up learning approaches such as reinforcement learning (RL), a team of agents can only learn emergent policies. However it may also be desirable to constrain policy search from top-down so that a team can learn more explicit policies in dynamic environments with continuous search spaces. In this paper we present a multiagent learning methodology that combines case-based learning and RL to address this need. Symbolic plans describe at a high-level the policies that a team of agents needs to learn for a wide variety of situations. For each high-level plan whose preconditions match the current state of their team, agents learn how to operationalize each step in that plan. For each training scenario, a team learns to find a sequence of actions that each agent in that team can execute such that each plan step can be operationalized under current external conditions; this application knowledge is acquired via RL. We use simulated robotic soccer to demonstrate this learning approach.