Neural Network Based Robot Navigation/ Motion Planning With 3d Mapping

Abstract

For environment monitoring without human intervention, autonomous 3D mobile robot mapping is widely utilized. But, poor performance has been shown by the prevailing techniques in the complex environment. Thus, to deal with this limitation, a novel framework named 3-Dimensional Mobile Robot Mapping and Motion Planning using Deep Q-Learning-based Markov Decision Model Deep Neural Network is proposed. Here, the primary sensors are utilized for robot navigation. Afterward, the point clouds are pre-processed and the similar pixels are grouped together; then, features are extracted. The Gazelle Optimization Algorithm (GOA) is utilized for enhancing the feature extraction phase. Next, the current posture of the robot is estimated by the Transformation matrix applied Single value decomposition Linear N-Point Camera Pose Estimation (TMSVDLCPE); also, grounded on the estimated pose, the desired view is captured. The captured images are then converted into 3D formats. The robot’s 3D images, speed, and current position are inputted to the DQMD-DNN, which efficiently plans the next optimal move of the robot. The experimental outcomes exhibited that the proposed technique withstands higher decision accuracy when contrasted with the prevailing frameworks