Purpose

In this paper, an attempt has been made to develop an algorithm equipped with geometric pattern registration techniques to perform exact, robust and fast robot localization purely based on laser range data.

Design/methodology/approach

The expected pose of the robot on a pre-calculated map is in the form of simulated sensor readings. To obtain the exact pose of the robot, segmentation of both real laser range and simulated laser range readings is performed. Critical points on two scan sets are extracted from the segmented range data and thereby the pose difference is computed by matching similar parts of the scans and calculating the relative translation.

Findings

In contrast to other self-localization algorithms based on particle filters and scan matching, the proposed method, in common positioning scenarios, provides a linear cost with respect to the number of sensor particles, making it applicable to real-time resource-limited embedded robots. The proposed method is able to obtain a sensibly accurate estimate of the relative pose of the robot even in non-occluded but partially visible segments conditions.

Originality/value

A comparison of state-of-the-art localization techniques has shown that geometrical scan registration algorithm is superior to the other localization methods based on scan matching in accuracy, processing speed and robustness to large positioning errors. Effectiveness of the proposed method has been demonstrated by conducting a series of real-world experiments.

中文翻译：

---

GSR：几何扫描配准算法，用于鲁棒和快速的机器人姿态估计

目的

在本文中，已经尝试开发一种装备有几何图案配准技术的算法，以完全基于激光测距数据来执行精确，强大和快速的机器人定位。

设计/方法/方法

机器人在预先计算的地图上的预期姿态为模拟传感器读数的形式。为了获得机器人的准确姿势，需要对真实激光范围和模拟激光范围读数进行分割。从分割的范围数据中提取两个扫描集中的关键点，从而通过匹配扫描的相似部分并计算相对平移来计算姿态差。

发现

与其他基于粒子过滤器和扫描匹配的自定位算法相比，该方法在常见的定位方案中相对于传感器粒子的数量提供了线性成本，使其适用于实时资源受限的嵌入式机器人。所提出的方法即使在非遮挡但部分可见的段条件下也能够获得对机器人相对姿势的合理准确的估计。

创意/价值

对最新定位技术的比较表明，几何扫描配准算法在精度，处理速度和对大定位误差的鲁棒性方面优于基于扫描匹配的其他定位方法。通过进行一系列实际实验证明了该方法的有效性。