【作者】 崔壮平；

【导师】 朱世强；

【作者基本信息】 浙江大学 ， 机械电子工程， 2011， 硕士

【摘要】 非结构环境下,移动机器人如何实现无碰撞的导航是当前研究的一大热点,也是一大难点。本文以吸尘机器人为研究对象,设计了基于该平台的多传感器硬件系统,形成了新的吸尘机器人,然后本文研究了基于该机器人的模糊神经网络避障算法,最后对基于该平台的定位算法做了一定的理论探索：首先,论文从实际角度出发,设计了一个新的多传感器硬件系统,并将新系统连接到吸尘机器人上取代以前的前视板,新系统包括15个超声波探头,可以测量14个方向5-50cm障碍的距离,然后在超声波探头中间插入了8对红外传感器,可以测量8个方向5cm以内的障碍物,这样,新的系统能保证机器人得到前半圈内大部分障碍物的距离和方位信息。同时,论文研究了各种超声波测距算法,以及超声波传感器和红外传感器测距的一些特性,使所设计得多传感器系统能得到最好的测量效果。其次,由于机器人在未知环境下避障,必然会需要一个合理的定位方法,故本文接下来在多传感器硬件系统的基础上对定位算法做了许多理论的探索,基于光电编码器的航位推测算法存在极大的累积误差,由此我们引入了扩展卡尔曼滤波算法融合光电编码器反馈数据和测距传感器的观测数据来校正,得到了非常满意的定位效果。但是扩展卡尔曼滤波的定位效果在系统非线性增强的情况下会变差,所以本文又采用了无迹卡尔曼滤波算法来定位,发现无迹卡尔曼滤波能很好的应用在非线性比较强的状态。最后,论文基于新平台做了许多相关避障算法的研究,提出了能应用于吸尘机器人的模糊神经网络模型,该模型以当前机器人障碍物距离信息、自身速度和自身相对目标角度为输入,以轮子的加减速度为输出,能比较好的控制机器人避开完全未知环境中的障碍,本文从理论出发,通过仿真确定模型的基本参数,然后根据需要采用神经网络方法训练调整参数,最后将训练好参数的模型移植到了吸尘机器人上面进行避障实验,实验中机器人能较好避开障碍物。更多还原

【Abstract】 Realizing non-collision navigation for mobile robot under the non-structured environment is currently a big hot spot, and is also a big difficulty. The dissertation take cleaning robot as the object of study, it designed a multi-sensor hardware system based on the cleaning robot platform and formed a new robot. Then a neurofuzzy-Based method was transplanted into the robot. Some theory explorations about localization were made upon our robot at last:Firstly, the dessertation designed a new multi-sensor from reality, and connected the new system to cleaning robot substitute the beforhand foresight board, the new system including 15 ultrasonic sensors, may survey the barrier distance of 5-50cm in 14 direction, then 8 pair of infrared sensors have been inserted in the middle of ultrasonic sensors, the infrared sensor can survey the obstacle of 5cm in 8 direction. Therefor, the new system can guarantee the robot to obtain the majority of obstacle distance and the azimuth information in the half-turn. At the same time, the dissertation has studied each kind of ultrasonic ranging algorithm, as well as the ultrasonic sensor and infrared sensor range finder’s some characteristics, making the multi-sensor systems be able to obtain the best survey effect.Secondly, the robot need a proper obstacle avoidance method since it work under a very complex environment, so the dissertation studied many localization algorithm based on multi-sensor hardware. Enormous accumulated error was existed in dead reckoning algorithm based on the electro-optical encoder, so we introduced EKF (Expanded Kalman Filtering) algorithm to fusion electro-optical encoder’s feedback data and the range finder sensor’s observation data to adjust the error, which can obtain very satisfactory localization effect. However, EKF method would get a worse effect in an environment with very strong nonlinear, which can be well substitued by UKF (Unscented Kalman Filtering) method.Finally, the dissertation did some research about obstacle avoidance algorithm based on the new platform, figured a neurofuzzy-based method which can be applicated on cleaning robot. The method takes the distance imformation of obstacle、 speed of itself and the angle between itself and goal as input, and wheel’s addition and subtraction speed as the output. It could make robot avoid abstacle quite well in unknown environment. The dissertation determined the basic parameter of the method through thoery simulation, and adjusted the parameter through neural network training. The module with certain parameter was finally transplanted to cleaning robot which can easily avoid obastacle in experiment.更多还原