Autonomous Quadcopter Navigation and Control

Robust Indoor Navigation and Object Detection using Stereo Camera and LiDAR

Abstract

This project involves the development of an autonomous navigation system for a quadcopter in simulated environments. The system integrates sensor fusion for localization, PID-based flight control, and dynamic path planning to enable the drone to traverse complex indoor environments. Leveraging the ROS 2 ecosystem and Gazebo simulation, the quadcopter performs autonomous navigation and collision-free movement.

1. Introduction

Autonomous aerial navigation is a critical component for applications ranging from indoor inspection to search and rescue. Unlike ground-based platforms, quadcopters must manage stability across six degrees of freedom while perceiving obstacles in 3D space. This project focuses on implementing an autonomous flight stack that coordinates flight dynamics with high-level mission planning using ROS 2.

2. System Overview

ROS 2 Humble: The software framework for managing asynchronous sensor data and control commands.
Gazebo Simulator: Used for physics-accurate flight simulation and sensor modeling (Lidar, IMU).
ArduPilot: Integration with flight control firmware for low-level motor stabilization and MAVLink communication.
Sensor Suite: Utilization of 2D Lidar and stereo camera for obstacle detection and IMU/Odometry for pose estimation.

Quadcopter Design — Fig. 1. Quadcopter platform used for simulation and testing

3. Methodology

3.1 Environment and Flight Setup

Design of a custom indoor environment in Gazebo featuring narrow corridors and dynamic obstacles.
Configuration of the quadcopter URDF/SDF models with optimized weight-to-thrust ratios.
Calibration of the simulated EKF (Extended Kalman Filter) for robust state estimation.

3.2 Stereo Camera Calibration

The stereo camera was manually calibrated using ROS2 nodes
A ROS2 node was developed and used in the pipeline for the image processing

3.3 Disparity Maps

The feature matching will be used for disparity maps
Depth maps will be computed from the disparity
The calibrated stereo camera was utilized

Feature Matching using stereo camera — Fig. 3. Feature Matching using a stereo camera

3.4 Cartographer SLAM

The LiDAR was used to generate Point Clouds for the SLAM algorithm
Real-time SLAM was utilized and the mapping was viewed via RViz.
The disparity map shown requires further tuning of the parameters for clearer map

SLAM Mapping and Dispiraity — Fig. 4. Cartographer SLAM and Disparity Map

3.5 Autonomous Navigation

In progress