Joint action optimation for robotic soccer multiagent using reinforcement learning method

摘要

In order to fulfill some tasks to reach a certain common goal, agents need to make sequence of decisions they have to perform as agroup. The decision is taken based on a selection mechanism of available actions. Choosing arbitrary action will lead to time and energy waste, since not all actions are even optimum. Agents need to decide not only which individual action that will lead to optimum performance, but also their joint actions. Applying reinforcement learning in the multiagent's learning process gives a sequence of optimum joint actions, which collaboration of agents based on this optimum joint actions guarantees the fastest time to reach their goal.