A practical implementation of navigation and obstacle avoidance for quadcopters

Abstract

Robotics is a multidisciplinary science field which includes the design, production and usage of robots. Robotic systems consist of robots and other devices and systems used with robots. Nowadays, robots are primarily used in the military, agriculture, food packaging and production sector. The robots that are used in these sectors increase productivity and automate the process, as well as work in areas that can be harmful to humans. The ability to be used in many areas and the surveillance capability of Unmanned Aerial Vehicles (UAV) increases its popularity, therefore increasing its usage in these sectors. UAVs can be defined as a type of aircraft with a thrust system, can be operated with a remote controller, and has a payload capacity. UAVs can be divided into two main categories: fixed-wing types and multi-rotor UAVs. Quadcopters can be described as multi-rotor UAVs with vertical takeoff capability. These UAVs use four actuators and navigate in a 3D coordinate system. These vehicles have six degrees of freedom which, three of them represents x,y and z axes and other three are rotation around these axes. The advancements in rare earth magnets made it possible to develop very lightweight but powerful BLDC motors. Advancements in battery technology enabled high power density batteries. With the help of MEMS technology, lightweight and fast sensors have become accessible, leading to quadcopters becoming a prevalent research topic. A Quadcopter consists of motors, ESCs, battery and autopilot hardware with sensors and a microcontroller to run flight control and navigation software. Flight control software runs on an embedded microcontroller and calculates the motor output by processing the sensor data and the reference input. In teleoperated systems, these inputs are generated by the operator via Ground Control Stations (GCS). In autonomous systems, high-level navigation software gives reference inputs to the flight controller, and flight control software gives sensor readings and quadcopter state to high-level navigation software. The software is designed in a way that operator can always take over the control of the aircraft in all flight modes in case of any failure or controller problem, ensuring the safety of the development platform. In this thesis, a quadcopter is built, and low-level and high-level software is developed. First, the necessary hardware, electronic modules and sensors for a quadcopter setup are determined. After building the development platform, sensor drivers and calibration software are developed for the NUCLEO-H7A3 development board. The gyroscope sensor is calibrated to get more accurate angular velocity readings. The accelerometer is calibrated and filtered to remove misalignment errors from the sensor data. The magnetometer sensor is calibrated to eliminate hard and soft iron effects. These calibration stages are crucial for the filtering algorithm. A robust quadcopter attitude estimation is achieved by fusing calibrated GPS, accelerometer, barometer and gyroscope readings. In order to control the quadcopter for attitude, position and velocity tracking, a cascaded PID controller structure is used. The attitude control of the quadcopter is achieved using six cascaded PID controllers, and the position control of the quadcopter is performed using seven cascaded PID controllers. In total, thirteen PID controllers are used in the flight control system. The controller performance is tested on the platform with different references. Online planning and replanning methods are commonly used in unknown or dynamic environments to ensure path safety. As high-level navigation and obstacle avoidance software, a novel local path replanning algorithm is proposed. The probabilistic Foam (PF) idea is adapted for reactive planning tasks. Probabilistic Foam infrastructure is selected for several reasons. Firstly, these types of planners are lightweight, and secondly, they ensure collision-free path generation. But they don't guarantee path length optimality. In this work, the latest achievement on the PF idea was further improved and adapted to local planning tasks. It is proved that this algorithm produces shorter paths and more lightweight than alternatives by conducting qualitative and quantitative tests. This algorithm is also tested as a local planner on coverage path tracking and 2D obstacle avoidance tasks designed for a quadcopter in a realistic simulation environment.