基于高斯牛顿的局部优化SLAM系统

Local Optimization SLAM System Based on GaussNewton Method

移动机器人的同时定位和地图重构一直是机器人研究的重要基本问题,有效地解决该问题被认为是真正实现移动机器人自主化的关键。激光测距仪的快速性以及抗噪性满足机器人导航需要的实时性和精确性,因此基于激光的同时定位和地图重构是实际中应用最广泛的方法。本文采用一种改进的快速获取占据栅格地图梯度的近似方法,利用Sobel算子作为相关核对栅格地图进行滤波,再进行双线性插值获得地图任意点的梯度值。通过高斯-牛顿方法来寻求每一帧新的观测数据对齐到现有地图的最优位姿,再根据位姿把观测数据更新到地图中,实验结果表明,改进方法可以实现更高精度的建图和定位的功能。

Simultaneous Localization and Mapping （SLAM） of mobile robots has been an important basicproblem in robotics research. Effective solution to SLAM problem is considered to be the key to realize the autonomy of mobile robots truly. The rapidity and noise immunity of laser rangefinder meet the real-time and accuracy of robot navigation. Therefore, the laser-based SLAM is the most widely applied method in practice. In this papcr, an improved method is used to get the gradient of occupying grid map quickly. Sobel operator is used to filter grid map and get bilinear interpolation to get the gradient value of any point. Gauss-Newton method is used to find the optimal pose of each new observation data to the existing map, and then the observation data are updated to the map according to the pose. The experimental results show that the improved method can achieve the functions of Localization and Mapping with higher accuracy.