论文总字数：33001字

摘 要

在移动机器人和自动驾驶领域，未知环境下的同时定位与建图（SLAM）问题是研究的热点问题。最近，采用改进建议分布和自适应重采样的Rao-Blackwellized粒子滤波算法有效地解决了同时定位与建图问题。该方法通过减少粒子数来提高运算效率，运用改进重采样方式来避免粒子贫化问题。同时，国内也出现了为降低成本，在Cortex-A53平台上实现激光雷达SLAM的系统，并完成了室内导航与避障的功能。

本文基于开源机器人操作系统（ROS）的分布式框架，用交叉编译的方法，在自主开发的，应用ARM与STM32为核心的机器人平台上实现了基于激光雷达的Rao-Blackwellized粒子滤波算法。通过学习粒子滤波算法，掌握了概率机器人基础理论知识，并熟悉了Linux的操作与ARM软件的开发。经过反复调试机器人的软硬件，最终在ARM平台上实现了高效且稳定的SLAM算法与自主导航，

在室内规则环境中进行了栅格建图与定位实验，并且将建立的栅格地图与搭载Kinect的TurtleBot所建地图进行比较。通过实验比较不同粒子数与重采样阈值的建图定位效果，得出在该环境下以上参数对结果的影响。另外，通过自主导航实验测试软硬件和算法的实现效果，分析了激光雷达作为传感器的优缺点与SLAM未来发展的展望。

关键词：同时定位与建图，Rao-Blackwellized粒子滤波，建议分布，重采样

Abstract

In the field of robotics and automatic driving, the problem of simultaneous localization and mapping (SLAM) in unknown environment is a hot research topic. Recently, Rao-Blackwellized particle filter algorithm with improved proposal distribution and adaptive resampling effectively solves simultaneous localization and mapping problem. By reducing the number of particles, the efficiency of calculation can be improved. Furthermore, the resampling method is improved to avoid the particle dilution problem. At the same time, in order to reduce the cost, radar-based SLAM is realized on a Cortex-A53 platform, and indoor navigation as well as obstacle avoidance is implemented in China.

By using the open source robot operating system (ROS), Rao-Blackwellized particle filter algorithm based on the laser radar is realized. Cross compiling method and the robot platform with ARM and STM32 as its core are involved. Through studying particle filter algorithm, the basic theory knowledge of probabilistic robot is mastered, as well as the operation of Linux and development of ARM software. After repeatedly debugging the hardware and software of the robot, an efficient and stable SLAM algorithm and autonomous navigation are realized on an ARM platform.

The grid mapping and localization experiments are carried out in the office compared with those from TurtleBot with Kinect. The effects of two parameters above are learnt by comparing the mapping results with different particle numbers and resampling thresholds. In addition, the implementation of autonomous navigation tests the hardware and software and relizes the algorithm. Both advantages and disadvantages of using the laser radar as a sensor are analyzed, which leads to a prediction to future SLAM.

KEY WORDS:SLAM, Rao-Blackwellized Particle Filtering,Proposal Distribution,Resampling

目 录

摘要 I

Abstract II

目 录 i

第一章 绪论 1

1.1 研究背景 1

1.2 国内外研究现况 3

1.3 论文研究内容 4

1.4 论文组织结构 4

第二章 ROS软件平台开发应用 5

2.1 引言 5

2.2 ROS开发框架 5

2.2.1 ROS发展背景和意义 5

2.2.2 ROS主要特点 6

2.2.3 ROS总体框架 6

2.3 ROS机器人 8

2.3.1 简介 8

2.3.2 本文实现载体 9

第三章 基于RBPF的未知环境定位与建图 11

3.1 建图问题 11

3.2 定位与建图分离 11

3.3 粒子滤波解决建图定位的问题 12

3.4 改进建议分布和自适应重采样 12

3.4.1 改进建议分布 12

3.4.2 自适应重采样 13

第四章 已知地图中定位与导航 14

4.1 引言 14

4.2 移动机器人定位 14

4.3 自主定位 14

4.4 全局路径规划 15

4.5 局部控制指令规划 15

第五章 实验设计与结果分析 17

5.1 实验设计 17

5.1.1 软件核心结构 17

5.1.2 交叉编译 17

5.2 实验步骤 18

5.2.1 搭载Kinect的TurtleBot实验 18

5.2.2 研发的机器人SLAM实验 18

5.2.3 研发的机器人自主导航实验 18

5.3 结果与分析 18

5.3.1 参考地图 18

5.3.2 粒子数与重采样的影响 21

5.3.3 自主导航 22

第六章 总结与展望 24

6.1 结论 24

6.2 实验反思与展望 24

致谢 25

参考文献 26

附录A 28

剩余内容已隐藏，请支付后下载全文，论文总字数：33001字