QQ客服 官方微博 反馈留言

题名	基于Spiking神经网络的移动机器人环境感知及行为控制的研究
作者	王秀青
学位类别	工学博士
答辩日期	2007-06-05
授予单位	中国科学院研究生院
授予地点	中国科学院自动化研究所
导师	谭民 ; 侯增广
关键词	移动机器人 Spiking神经网络 自主导航 多传感器信息融合 分类 Mobile Robot Spiking Neural Networks (SNNs) Autonomous Navigation Multisensor Information Fusion Classification
其他题名	Research on Environment Perception and Behavior Control for Mobile Robot Based on Spiking Neural Networks
学位专业	控制理论与控制工程
中文摘要	随着科学技术的发展，移动机器人技术也进入了新的发展阶段，机器人的智能水平正向更高的层次发展。神经网络作为智能技术的重要方面也得到了长足的发展，并出现了第三代人工神经网络——Spiking神经网络。由于Spiking神经元模型更接近真实生物神经元，使得Spiking神经网络具有前两代神经网络所不可比拟的优点，因而引起了神经网络研究领域的广泛关注。本文结合Spiking神经网络，对移动机器人行为控制和环境感知问题进行研究，本文工作的主要贡献体现在： 提出基于多超声传感信息的Spiking神经网络避障行为控制器的设计，并对相应的实验结果进行了分析、讨论。该控制器中神经网络的训练采用基于Spike的Hebb学习规则，使机器人可以在线学习。实验结果表明：该控制器结构简单，易于实现，可以实现有效的避障。 提出基于Spiking神经网络的墙壁跟踪行为控制器的设计，并进行了仿真实验。实验结果证明了墙壁跟踪控制器的有效性。在避障控制器、墙壁跟踪控制器的基础上，设计了无碰撞趋向目标点的复合行为导航，仿真实验验证了所设计的导航控制器的合理性。 在多超声传感器测量信息融合的基础上，研究了基于Spiking神经网络，主元分析、核主元分析，基于Levenberg-Marquardt 算法的BP网络等智能计算方法在移动机器人环境感知问题上的应用。针对常见的七种走廊场景，成功地设计了基于上述方法的走廊场景分类器，并从分类效果、方法鲁棒性等方面进行了分析、比较。基于核PCA和Spiking神经网络的走廊场景分类方法均具有较好的分类效果和较强的鲁棒性。
英文摘要	With the development of science and technology, the research for mobile robot has reached a new stage, and the intelligent level of the robot has been improved greatly. Neural Networks have developed greatly, and spiking neural networks (SNNs) - the third generation of artificial neural networks (ANNs) appeared. SNNs have advantages over the previous two generations of ANNs in many aspects, because they have more plausible neural models than the previous ones. SNNs have attracted great attentions in neural networks research area. This dissertation focuses on mobile robot's behavior control and environmental perception, which are based on SNNs. The main contributions of this dissertation are as follows: Using multi-sonar-sensor-information, the design of the behavior controller based on SNNs for mobile robot to avoid obstacles is proposed. The structure of the controller is easy to implement and the SNN in the controller can be tuned on line by spike-based Hebbian learning ruler. The simulation results are analyzed and discussed, which show the controller is effective. The wall-following controller based on SNNs is also proposed, and the simulation results show that the robot can follow walls very well by the controller. Combined with the obstacle-avoidance controller and the wall-following controller, which are based on SNNs, the navigation controller is designed. The navigation simulations involving composite behaviors such as obstacles avoidance, wall trace and the objective point approximation, verified the effectiveness of the sensor-based autonomous navigation controller of mobile robot in unknown and unstructured environments. With multi-sensor information fusion, such intelligent computing methods as: SNNs, PCA, Kernel-PCA, and BP NNs, are applied in mobile robot's environmental perception. Using the above methods, classifiers for the seven commonly corridor scenes are designed. The classifying results are then analyzed and compared. The classifiers based on Kernel-PCA and SNNs all have higher corridor scenes recognization rate and better robustness.
语种	中文
其他标识符	200418014628040
内容类型	学位论文
源URL	[http://ir.ia.ac.cn/handle/173211/6001]
专题	毕业生_博士学位论文
推荐引用方式 GB/T 7714	王秀青. 基于Spiking神经网络的移动机器人环境感知及行为控制的研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2007.

个性服务

查看访问统计

相关权益政策

暂无数据

收藏/分享

除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。