Heuristic reinforcement learning for symmetric and asymmetric multi-agent systems
Chunyi Shi · Journal of Tsinghua University(Science and Technology) · 2006
In traditional multiagent reinforcement learning,an agent can only learn by itself and a heuristic algorithm without multiple.This problem was solved by multiagent reinforcement heuristic-based learning.In this system agents communicate between each other to get information and to select strategies,with heuristics to cooperate with each other,and to improve the efficiency of learning.Tests with 2 agents,2 status,3 actions showed that the method converges faster than traditional distributed reinforcement learning.