Autonomous Navigation Of UAVs In Complex Indoor And Outdoor Environments

The integration of Unmanned Air Vehicles(UAVs)– also commonly referred to as drones – into our society has been growing at an alarming pace.The number of industries that use drones is exponentially growing.UAVs are also attracting increased interest from various communities,such as defence,emergency response,disaster relief,healthcare,agriculture,mining,infrastructure development,sports,education,and many others.This phenomenal growth in the use and interest of UAVs has its downside as well: The growing use of UAVs brings up numerous challenges,such as reacting optimally to dynamically changing,unseen and unstructured scenarios;unreliable state estimation;coupling actions and perceptions in real-time under severe resource constraints,to name a few.In addition,robotics systems in general and UAVs face several major problems simultaneously,such as perception,control,path planning,and localization,especially when operating in a GPS-denied environment.This research dissertation is an attempt to encapsulate the advantages as well as challenges that accompany the increased use and integration of U AVs into our society.The dissertation adds to the existing literature some recommendations that are believed to enhance the operational capabilities of the autonomous UAVs under external disturbances,both indoor and outdoor environments.The dissertation starts with presenting Event-driven programming-based path planning and navigation of UAVs around a complex urban environment.The algorithms used are A* search algorithms to voxel map,probabilistic roadmap,and Rapid Exploring Random Tree 2D grids to 3D motion planning using random sampling,heuristic,collinear,and path pruning methods in 2D and 3D realistic environments.To control UAVs amidst external disturbances,like wind gust,the dissertation proposes disturbance rejection based optimized controllers,where a Two Degree-of-Freedom Proportional Integral Derivative(2DOF-PID)controller with Particle Swarm Optimization(PSO)is proposed for position controller,while attitude controller is based on Robust Adaptive Integral Backstepping(RAIB).Assuming the knowledge of the predetermined limits of the external and unstructured disturbances,a guaranteed quality of transient and stead y-state tracking performance is obtained.In addition,to optimize the power consumption in the presence of strong wind gusts,the Blade Element Momentum Theory(BEMT)model is used along with the proposed control design,which reduces the power consumption while keeping the UAVs on the desired track.Next,the dissertation proceeds to assess the development of perception-action aware-based autonomous drone race in a photorealistic environment.Autonomous drone racing presents a research challenge at the intersection of computer vision,path planning,state estimation,and control.The dissertation proposes an approach that integrates a deep convolutional neural network(CNN)with state-of-the-art path planning,state estimation,and control algorithms to confront this challenge.To this end,a novel and computationally efficient gate-detection and VINS-Mono method are implemented to detect the corners of gates and get information about the motion of the quadrotor.Then a more efficient EKF is implemented t o fuse this information to get the gate’s map and quadrotor’s state.The planner and controller then use this information to generate a short,minimum-snap trajectory segment and send corresponding motor commands to reach the desired goal.This approach outperforms state-of-the-art methods and flies more consistently than many human pilots demonstrated by extensive experiments.In addition,the proposed system appears to successfully guide the drone through tight race-courses reaching speeds up to 7m/s,demonstrated and recognized during the 2019 competition of Alpha-Pilot Challenge,which was organized jointly by Lockheed Martin and Drone Racing League.The dissertation then delves into acquiring a safe and secure autonomous trajectory tracking of UAVs,especially in a highly dynamic environment,and proposes Ka NET,a convolution neural network that can safely drive and improve tracking performance.Simultaneously cascaded PID+FF(Feed-Forward)algorithms are used to control the attitude/altitude of a flying vehicle.Efficient implementation of Ka NET-based PID+FF method controls flying vehicles through complex environments and cooperation with nearby decision-making agents(e.g.,buildings,other flying vehicles).Finally,a Deep Reinforcement Learning(DRL)and safe DRL with adaptive backstepping based Linear Parameter Varying Model Predictive Control(LPV-MPC)are presented in this dissertation.This research dissertation first presents reviews,then proposes DRL methods in the context of path planning,na vigation,and control of UAVs.Then,a safe DRL method is developed to follow the safe path generated live from the onboard camera or other sensors,or it can be a programmed track.Finally,adaptive backstepping-based LPV-MPC controllers are proposed to control the position and attitude of UAVs.