基于多传感器融合的室内机器人自主导航方法研究

发布时间：2018-02-20 00:38

本文关键词： 室内机器人自主导航多传感器融合里程计修正方向A~* 场景分类　出处：《中国科学技术大学》2017年博士论文　论文类型：学位论文

【摘要】：自主导航是室内机器人实现自决行为规划的核心问题,相对复杂条件下的环境感知,基于多传感器融合成为新的发展趋势。本课题以机器人平台为工程应用背景,期望将深度学习引入人工智能领域,开展相关机器人技术在自主导航中的方法研究,开发集机械平台、嵌入式硬件、软件系统、SLAM算法、场景识别方法于一体的机器人综合系统框架,结合多传感器融合的环境信息,实时指导路径规划,不仅提供室内机器人自主导航的可行方法,对促进该领域语义地图的发展也具有一定积极意义。本文主要内容如下:研究并设计满足自主导航条件的机器人综合系统框架,包括集主动感知、语音输入、运动控制等的机械平台,集IO驱动、数学运算库、机器人控制、运动学和动力学模型等的嵌入式硬件结构,集环境感知与执行、移动终端、中间层连接、分布式架构、任务算法等的机器人系统。研究适用于ROS环境的多传感器融合SLAM算法:融合超声、RGBD、激光雷达数据,基于Gmpping建立二维栅格地图,通过自适应蒙特卡洛定位,导航时采用A*算法实施全局探索,并辅以动态窗口法进行局部避障。实验证明所创建地图分辨率为2cm,并能指导机器人进行室内导航。提出一种混合IMU的里程计实时校正算法:基于扩展卡尔曼滤波和互补滤波融合获取姿态角,修正机器人方位信息;定时比较里程计位移与二次积分加速计数据,防止机器人运动漂移及悬空。相关传感器经过零漂校准后,融合实验数据精度高、收敛速度快,并能校正里程计丢失现象。提出一种用于路径规划的方向A*算法:首先采用“视野线”平滑原则优化路径,消除锯齿效应并避免部分碰撞;其次应用“圆弧-直线-圆弧”转弯策略,避免机器人本体宽度影响;最后基于二叉堆加速算法,提升算法计算效率。仿真实验结果表明,方向A*算法满足平滑要求且能有效避免碰撞,其加速方法可平均提速4-7倍。同时,机器人在真实实验环境下实现安全自主导航,跟踪误差小于0.15m,验证了方法的可行性。提出一种基于语义建图的场景分类方法:首先映射深度信息构建二维栅格地图,自主规划场景识别路径;其次基于卷积网络建立场景分类模型,实时识别脱离特定训练;最后利用贝叶斯框架融合先验知识,修正错误分类并完成语义建图。实验结果表明,机器人能够进行全局自主探索,实时判断场景类别,并创建满足要求的三维语义地图。同时,在实际路径规划中,机器人可以根据语义信息改善导航行为,验证了方法的可行性。
[Abstract]:Autonomous navigation is the core problem of indoor robot's self-determination behavior planning. Environment perception based on multi-sensor fusion has become a new development trend under relatively complex conditions. In this paper, the robot platform is used as the engineering application background. It is expected that the deep learning will be introduced into the field of artificial intelligence, the research on the methods of robot technology in autonomous navigation will be carried out, and the integrated mechanical platform, embedded hardware and software system will be developed, and the slam algorithm will be developed. The integrated robot system framework of scene recognition method, combined with multi-sensor fusion of environmental information and real-time guidance path planning, not only provides a feasible method for indoor robot autonomous navigation. The main contents of this paper are as follows: the research and design of robot integrated system framework which meets the conditions of autonomous navigation, including set active perception, voice input, Mechanical platform for motion control, integrated IO drive, mathematical operation base, robot control, kinematics and dynamics model embedded hardware structure, environment sensing and execution, mobile terminal, intermediate layer connection, distributed architecture, etc. Research on multi-sensor fusion SLAM algorithm suitable for ROS environment: fusion ultrasonic data, lidar data, build two-dimensional raster map based on Gmpping, and adopt adaptive Monte Carlo localization, In navigation, we use the A* algorithm to implement the global exploration. Experimental results show that the proposed map has a resolution of 2 cm and can guide the robot to conduct indoor navigation. A real-time correction algorithm for mileage based on hybrid IMU: extended Kalman filter (EKF) is proposed. And complementary filtering fusion to obtain attitude angle, To correct the azimuth information of the robot, to compare the displacement of the odometer with the data of the second integral accelerometer, to prevent the robot movement drift and suspension. After zero drift calibration, the fusion experiment data has high accuracy and fast convergence speed. A direction A * algorithm for path planning is proposed. Firstly, the "field of view" smoothing principle is used to optimize the path to eliminate the sawtooth effect and avoid partial collision. Secondly, the turning strategy of "circular arc, straight line and circular arc" is applied to avoid the influence of robot body width. Finally, based on the binary stack acceleration algorithm, the computational efficiency of the algorithm is improved. The simulation results show that, The directional A * algorithm meets the requirements of smoothing and can effectively avoid collision. Its acceleration method can increase the speed by an average of 4 to 7 times. Meanwhile, the robot realizes safe autonomous navigation in real experimental environment. The tracking error is less than 0.15 m, which verifies the feasibility of the method. A scene classification method based on semantic mapping is proposed. Firstly, a two-dimensional raster map is constructed with mapping depth information, and the scene recognition path is planned independently. Secondly, the scene classification model based on convolutional network is established to recognize and separate from the specific training in real time. Finally, the Bayesian framework is used to fuse the prior knowledge, correct the error classification and complete the semantic map. The experimental results show that, The robot can make global autonomous exploration, judge scene category in real time, and create 3D semantic map that meets the requirements. Meanwhile, in actual path planning, robot can improve navigation behavior according to semantic information. The feasibility of the method is verified.
【学位授予单位】：中国科学技术大学
【学位级别】：博士
【学位授予年份】：2017
【分类号】：TP242

【相似文献】