A LEARNING AGENT BASED ON REINFORCEMENT LEARNING

Reinforcement learning can find optimal behavior sequence and perform on-line learning in dynamic environments by means of non-decisive rewards. Therefore reinforcement learning is one of the basic technologies of intelligent agent. The BDI model is extended, bringing strategy and capability into mental state, and strategy construct function is studied by means of reinforcement learning technology in an agent oriented development environment for intelligent software system, named AODE. An adaptive agent based on reinforcement learning is brought forward and a strategy construct function is realized. In this paper, an adaptive agent’s architecture is studied deeply so that agent is able to learn on-line in dynamical surroundings and satisfy every mental requirements of agent effectively.