Improved CNN-Based Path Planning for Stairs Climbing in Autonomous UAV with LiDAR Sensor

Abstract

Unmanned aerial vehicles (UAV) technology has been an innovative advancement in the scientific environment over recent years. In this paper, we propose an approach that facilitates UAVs with a monocular camera combined with light detection and range (LiDAR) sensor to navigate autonomously for stairs climbing in completely unknown, GPS-denied indoor environments. The suggested approach utilizes a state-of-the-art CNN model for the task. We suggest a novel approach utilizing the video feed derived from the UAV front camera to determine the next maneuver in the deep neural network model. The process is viewed as a classification activity, where the deep neural network model classifies the image as a stair or no-stair and LiDAR sensor data are used for distance calculation. The training is performed from a dataset of images obtained from multiple stairs. We show the effectiveness of the proposed device in indoor stairs scenarios in real-time.