A Multi-agent Cooperative Reinforcement Learning Algorithm Based on Team Markov Game
Li Yao · Fudan xuebao. Ziran Kexue ban · 2004
The research of learning behavior in multi-agent system is very important for adaptability of intelligent systems. It aims at the learning process of a kind of cooperative teams, which pursue the maximum benefit of a whole system. A new cooperative reinforcement learning algorithm based on Markov Game is proposed. Each agent of the team decides its behavior after forecasting the behavior strategy of acquaintances according to Markov Game structure and adapting appropriate exploring strategy. A jointly optimal behavior strategy will be reached through the learning process. An experiment is presented to show that the algorithm is effective.