基于卷积神经网络的移动机器人二维可定位性评估方法

A localizability estimation method based on convolutional neural networks for mobile robot

定位是移动机器人领域的关键技术,也是实现自主移动等各种任务的基础。可定位性是对给定位置实现准确定位能力的度量,是进一步在上层任务中避免定位失败的重要指标。基于提高机器人定位准确率的目的,采用信息融合理论与卡尔曼滤波理论,建立数学模型对基于激光雷达的地图匹配定位方法进行定位误差分析,明确了可定位性的主要影响因数。采用卷积神经网络设计可定位性熵估计器估计机器人定位熵,通过室内外移动机器人可定位性试验,得出本文方法在给定地图中对机器人定位熵估计值与实测定位熵值相比误差约在6%左右,能够对机器人可定位性进行准确估计。

Localization is one of the key technologies of mobile robot and a foundation of other tasks like autonomous motion etc. Localizability is the evaluation of the capability of localization,which also implies the probability of failure in localization. With the purpose of improving the accuracy of localization of robot,the information fusion theory and Kalman filter theory are used to establish a mathematical model to analyze the error of the lidar-based localization method,and the factors mainly affect the localizability are clarified. The convolutional neural network is used to design a localization entropy estimator to estimate the robot’s localization entropy. The tests of the localizability estimation in both indoor and outdoor environments show that the estimation error is about 6%,which proves that the proposed method can estimate the localizability accurately.