文章摘要
赵辰豪, 吴德伟, 何晶, 韩昆, 来磊.基于改进Q学习算法的导航认知图构建[J].空军工程大学学报:自然科学版,2020,21(2):53-60
基于改进Q学习算法的导航认知图构建
Navigation Cognitive Map Construction Based on Improved Q-Learning Algorithm
  
DOI:
中文关键词: 类脑导航  网格细胞  位置细胞  改进Q学习算法  导航认知图
英文关键词: brain-like navigation  grid cell  place cell  improved Q-learning algorithm  navigation cognition map
基金项目:国家自然科学基金(61603409)
作者单位
赵辰豪, 吴德伟, 何晶, 韩昆, 来磊 空军工程大学信息与导航学院西安710077 
摘要点击次数: 0
全文下载次数: 0
中文摘要:
      针对导航认知图构建效率低,方向信息不准确等问题,提出了一种基于改进Q学习算法的导航认知图构建方法。首先,利用径向基(RBF)神经网络学习生成网格细胞到位置细胞的映射关系,并利用位置细胞对空间进行表征;其次,采用改进Q学习算法学习位置细胞面向目标的Q值大小;最后,根据重心估计原理计算面向目标的方向信息,并生成导航认知图。仿真结果表明:与传统Q学习算法相比,文中算法生成导航认知图的学习次数从2 000次缩减至1 000次,提高了导航认知图的构建效率;学习值(指面向目标的方向信息)的相对误差最大降低了15%,提高了认知图的准确性。
英文摘要:
      In order to improve the efficiency in navigation cognitive mapping and reduce the error of direction information, a method of navigation cognitive mapping based on improved Q-learning algorithm is proposed in this paper. Firstly, Radial Basis Function (RBF) neural network is utilized for trainning the mapping relationship between grid cells and place cells, and the place cells are used to convey space information. Secondly, the improved Q-learning algorithm is used to learn the target oriented Q-value of the place cell to obtain direction information towards target. Finally, the center of gravity estimation principle is used to generate the target oriented direction information, constructing the navigation cognition map. The simulation results show that learning rounds of the navigation cognitive mapping generated by this algorithm is reduced from 2 000 to 1 000 compared with the traditional Q-learning algorithm, improving the construction efficiency of navigation cognitive map. Meanwhile, a maximum reduction of 15% in relative error is achieved, improving the precision of navigation cognitive mapping.
查看全文   查看/发表评论  下载PDF阅读器
关闭