Indoor obstacle avoidance system design And evaluation using deep learning and SLAM based approaches

Abstract

This project aims to contribute to the vibrant fiel do fobstacl edetectio nan dsaf eau-tonomous navigation by designing a robust and cost-efficien tsyste mfo rindoo rmobile robot obstacle avoidance. The system combines 2D LiDAR-based SLAM with the state-of-the-art RRT algorithm for effective path planning. In addition, a pioneering deep learning approach addresses challenges in SLAM-RRT-based obstacle avoidance, including un-certain sensor measurements, complex environments, generalization, planning efficiency, and non-geometric information. The deep learning model is trained using data from a simulated environment with a 2D LiDAR sensor, serving both SLAM and data acquisition purposes. Comparative analysis between odometry-based and SLAM-based pose compu-tation methods provides insights into successful deep learning-based obstacle avoidance. Implemented within ROS2, this project represents a significan tstrid ei nexplorin gcutting-edge techniques for robust and cost-efficien tindoo rmobil erobo tobstacl eavoidance