【摘要】 多机器人环境中的学习,由于机器人所处的环境是连续状态,连续动作,而且包含多个机器人,因此学习空间巨大,直接应用Q学习算法难以获得满意的结果。文章研究中针对多智能体机器人系统的学习问题,提出自适应模糊RBF神经网络强化学习算法,网络本身具有模糊推理能力、较强的函数逼近能力以及泛化能力,因此,实现了人类专家知识与机器学习方法的结合,减少学习问题的复杂度;实现连续状态空间与动作空间的策略学习。更多还原

【Abstract】 The learning in the multi-robots undertaking the team task in the dynamic enviroment is studied.Since the enviroment state and action is continous,and involing multi-robots,the learning space is huge,in finite state spaces,it is impossible to exactly store the optimal Q value function with lookup table representations,so it is difficult to use Q-learning directly.This thesis puts forward a reinforcement learning on the basis of adapted fuzzy RBF neural network with Q-learning to mapping from the state space to the action space.This method can build an ANFIS according to ex-perts’ knowledge,and be able to adjust parameters of the system’s antecedents and consequents in a self-adapt way.So it can establish a correct map to describe the cooperation among the robots.更多还原