论文部分内容阅读
针对复杂未知环境下难以获得完善的模糊导航控制规则以及传统的强化学习算法不能解决连续状态空间和连续动作空间的学习问题,提出了一种模糊强化学习算法.通过将模糊推理系统和强化学习算法相结合,设计了一种模糊强化学习系统,一方面,在缺乏专家经验的情况下,利用强化学习中的Sarsa(λ)学习算法来获取模糊逻辑控制器的模糊规则库,另一方面,利用模糊推理系统所具有的广泛逼近性,使机器人在学习时可以遍历到每一个状态动作对.同时将有限的专家经验引入到模糊推理系统,使Sarsa(λ)学习具备一定的先验知识,从而加快学习速度.仿真实验表明,该方法具有较好的实时性和鲁棒性,能够有效解决移动机器人在未知复杂环境中的导航问题.
Aiming at the problem that it is difficult to obtain perfect fuzzy navigation control rules in complex unknown environment and the traditional reinforcement learning algorithm can not solve the learning problems of continuous state space and continuous motion space, a fuzzy reinforcement learning algorithm is proposed. By combining fuzzy inference system and reinforcement learning algorithm A kind of fuzzy reinforcement learning system is designed. On the one hand, Sarsa (λ) learning algorithm in reinforcement learning is used to obtain the fuzzy rule base of fuzzy logic controller in the absence of expert experience. On the other hand, The extensive approximation of fuzzy inference system enables the robot to traverse each pair of state actions while learning, and introduce limited expert experience into the fuzzy inference system to make Sarsa (λ) learning possess a certain priori knowledge Speed up the learning speed.The simulation results show that this method has good real-time and robustness and can effectively solve the problem of mobile robot navigation in unknown complex environment.