|
Learning Approach to Coordinate Exploration with Limited Communication in Continuous Action Games Host Publication: the Adaptive and Learning Agents Workshop Authors: A. Rodriguez, P. Vrancx, R. Grau and A. Nowé Publisher: IFAAMAS Publication Year: 2012 Number of Pages: 7
Abstract: Learning automata are reinforcement learners belonging to the category of policy iterators. They have already been shown to exhibit
nice convergence properties in discrete action games. Recently, a
new formulation for a Continuous Action Reinforcement Learning
Automaton (CARLA) was proposed. In this paper we study the behavior of these CARLA in continuous action games and propose a
novel method for coordinated exploration of the joint-action space.
Our method allows a team of independent learners, using CARLA,
to ?nd the optimal joint action in common interest settings. We ?rst
show that a set of agents using CARLA will converge to a local optimum in a continuous action game. Using this property, we then
introduce a method for coordinating exploration to ?nd the global
optimum of the game. We also validate our approach in a number
of experiments.
|
