Convolutional Neural Network based Sensors for Mobile Robot Relocalization

Abstract

In this paper, we present a robust and real-time convolutional neural network (CNN) based localization algorithm that can be implemented on cheap and low power computation devices, thereby making it more practical. The proposed method first trains a CNN that takes RGB images from a monocular camera as input and performs regression for robot pose. It then incorporates the output of the trained CNN in an Extended Kalman Filter (EKF) for robot localization. Here we present a complete framework for localization of mobile robots in GPS-denied indoor and outdoor environments that do not require costly and computationally expensive sensors for real-time deployment. We demonstrate the performance of the proposed approach using a low power GPU, NVIDIA Jetson TX1, mounted on an indoor and an outdoor mobile robot platform. We show that the proposed approach gives promising results with localization error of 0.3 m in indoor environments and 1.3 m in outdoor environments. This makes CNN-sensors real-time, robust, low-cost and less computation-intensive substitutes for other sensors, with immense potential for a wider use in mobile robotics.