| 题名 | 基于Spiking神经网络的机器人控制研究 |
| 作者 | 王旭 |
| 学位类别 | 工学博士 |
| 答辩日期 | 2011-05-26 |
| 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 |
| 关键词 | Spiking神经网络 目标跟踪 多机器人系统 队形控制 围捕 图像信息提取 目标识别 spiking neural network target tracking multi-robot system formation control hunting image information extraction object recognition |
| 其他题名 | Research on Robot Control Based on Spiking Neural Networks |
| 学位专业 | 控制理论与控制工程 |
| 中文摘要 | Spiking神经网络采用脉冲的形式来传递和处理信息，可以充分利用输入信号的时间和空间信息，抗干扰能力强，易于硬件实现，适合于动态环境下的机器人控制。本文开展基于Spiking神经网络的机器人控制研究，主要内容如下： 首先，本文综述了移动机器人和多机器人系统的研究现状，同时描述了Spiking神经网络的基本特征及其在机器人控制中的应用，并对研究背景和论文结构做了介绍。 其次，设计了基于Spiking神经网络的移动机器人目标跟踪控制方法。机器人通过视觉、码盘以及超声传感器获取目标信息及障碍信息，预处理后送入神经网络进行脉冲编码、融合，与电机对应的正/反向神经元相互竞争生成神经网络的输出信号，滤波限幅后送入电机中，驱动机器人朝目标无碰运动。 第三，提出了一种基于Spiking神经网络的领航-跟随机器人队形控制器。网络分三层，分别采用LIF模型、近似一致性编码以及SRM模型，完成对传感器和任务相关信息的编码、融合以及控制信号生成，最终实现无碰的队形控制。对采用一致性编码的神经元的点火率进行了分析，并引入参考信号对输入信号相关性进行了检测。 第四，设计了基于模块化Spiking神经网络的多机器人围捕控制器，采用延时编码，由12个神经元模块和4个电机神经元组成。每个模块编码、融合来自其敏感方向周围的目标、传感及协调相关信息，输出信号送入电机神经元，进而实现对动态目标的围捕，网络权值采用基于随机策略的Hebbian学习进行调整。还针对采用延时编码的神经元，分析了噪声对其输出信号可能造成的影响。 第五，设计了一个递归的三层Spiking神经网络用于图像信息提取。输入层神经元具有输入/输出方向选择性，提取边缘信息；中间层神经元融合输入层的输出脉冲，并反馈到输入层；输出层浓缩中间层提取的图像信息。在神经元局部连接的情形下，该Spiking神经网络能够提取物体的位置以及大小等信息。通过模版匹配，对有颜色信息目标识别和无颜色信息目标识别进行了研究。 最后，论文对所取得的研究成果进行了总结，并阐述下一步的工作。 |
| 英文摘要 | The spiking neural network (SNN) encodes and processes the information inthe form of spikes to take full advantage of temporal and spatial information with high interference resistance and easy hardware realization. It is suitable for the robot control in dynamical environments. This thesis focuses on the research of robot control based on SNN. The contents are as follows: Firstly, the research development of the mobile robot and multi-robot systems are given. Meanwhile, the characteristics of SNN and its application to the robot control are presented. The structure of this thesis are also introduced. Secondly, a SNN based autonomous robot controller is designed for target tracking. The controller encodes the pre-processed environmental and target information provided by CCD cameras, encoders and ultrasonic sensors into spike trains, which are integrated by the SNN. The outputs of SNN are generated based on the competition between the forward/backward neurons corresponding to each motor, and then they are sent to the motors after filtering and restriction, by which the robot can move towards the target without collisions. Thirdly, a SNN based controller is designed to fulfill the task of formation control of multiple mobile robots by using the leader-follower strategy. The neural network contains three layers: the input layer encodes the inputs including sensor and task-related information by leaky integrate-and-fire (LIF) neurons, the hidden layer uses the approximate coincidence detection coding to fuse the information from the input layer and the spike response model (SRM) is applied to the output layer to fire spikes to drive the motors. The firing rate of the neuron based on temporal coincidence coding is estimated. Also, a reference signal is introduced to detect the correlation between inputs. Fourthly, a robot controller based on modular SNN is proposed for the coordinated hunting of multi-robot system with time-to-first-spike coding. The controller utilizes 12 direction-sensitive modules to encode and process the inputs including the environment, target and coordination information, and then four motor neurons are used to generate the motor control signals. The Hebbian learning with a stochastic form is applied to adjust the connection weights. Also, the output disturbance caused by the noise is analyzed. Fifthly, a three-layer recurrent spiking neural network with local connections is designed to extract the image info... |
| 语种 | 中文 |
| 其他标识符 | 200818014628018 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/6341]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 王旭. 基于Spiking神经网络的机器人控制研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2011. |
| 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。
修改评论