A Primer on Autonomous Aerial Vehicle Design

Hugo Coppejans, Herman Myburgh
2015 Sensors  
There is a large amount of research currently being done on autonomous micro-aerial vehicles (MAV), such as quadrotor helicopters or quadcopters. The ability to create a working autonomous MAV depends mainly on integrating a simultaneous localization and mapping (SLAM) solution with the rest of the system. This paper provides an introduction for creating an autonomous MAV for enclosed environments, aimed at students and professionals alike. The standard autonomous system and MAV automation are
more » ... iscussed, while we focus on the core concepts of SLAM systems and trajectory planning algorithms. The advantages and disadvantages of using remote processing are evaluated, and recommendations are made regarding the viability of on-board processing. Recommendations are made regarding best practices to serve as a guideline for aspirant MAV designers. The largest problems that MAVs face are the stabilization and control in six degrees of freedom (DoF), more commonly referred to as attitude and position control. The attitude control can been solved using a simple proportional-derivative (PD) controller. Other techniques can also be applied, such as "sliding mode" and "backstepping" [5, 6] . The second problem of position control is much more complex and will be the focus of this paper. Without position control on an MAV, it will be prone to drift because of the constant corrections the attitude controller has to make. The MAV will also be unable to localize itself within a known environment. This problem can easily be solved using GPS [7] and has been the preferred method for UAV designs in the past. For MAVs, however, GPS is not a reliable service, as it is prone to lose accuracy and connection in urban canyons and indoor areas. Because of this, an alternative solution must be found for MAVs. Some of the earliest methods involve the use of externally-mounted cameras (i.e., off board) to track the movement of the platform. The platform is mounted with reflective tags that the cameras pick up, and by knowing the exact position of the cameras, the platforms position can be inferred [8, 9] . This method has numerous disadvantages: the MAV is limited to areas that are covered by the cameras, and the cameras need manual installation and calibration. Thus, this method is limited to areas that are accessible by humans. The only alternative to this method is to use vision sensors that allows the platform to capture the environment surrounding it. This allows the platform to freely explore without a constraint on the area in which it can operate. The difficulty arises when the MAV needs to build a map of its surroundings and then localize itself within that map. A method called simultaneous localization and mapping (SLAM), coined by D. Rye, H. Durrant-Whyte and E. Nebot [10], is the preferred method for solving this problem. SLAM is the problem of placing a mobile robot in an unknown environment (no prior knowledge) wherein the robot must then be able to incrementally build a consistent map of this environment while simultaneously determining its own location with regard to the map that is being built [11] . The concepts of SLAM have been defined and tested, but the practical realization of these concepts is still an ongoing field of research. Recently, great strides have been made in creating MAVs that are able to operate in unknown, cluttered and GPS-denied environments using SLAM [12] [13] [14] . While the problem has been solved to a point where it can be used in a practical application, constant research is being done in order to create alternative solutions or to make improvements. More advanced problems, such as autonomous exploration, swarm navigation and large trajectory planning, are also being researched [14] [15] [16] . This paper discusses a quadcopter system, tasked with mapping an entire indoor environment. Because the final system consists of so many complex systems that need to be integrated perfectly, this paper presents a complete discussion of all of the core concepts and subsystems and how each subsystem is integrated into the final system. A Microsoft Kinect is the sensor of choice, but alternatives and trade-offs are discussed. This paper focuses on the most recent techniques that are available. Since the choice of hardware is of chief importance, an entire section is dedicated to discussing the hardware requirements. This paper is structured as follows: The full system design is described in Section 2, while the hardware requirements for the quadcopter platform are discussed in Section 3. The Microsoft Kinect and alternative sensors are evaluated in Section 4, and Section 5 states the advantages and disadvantages of using remote processing. In Section 6, probabilistic SLAM is defined, and a few core SLAM algorithms are discussed. The importance of trajectory planning is discussed in Section 7, and Section 8 aims to provide recommendations regarding the best techniques to use and also to give advice on how to approach the problem of creating an autonomous quadcopter of this kind. System Design Because MAVs are considered to be highly unstable and non-linear systems [13] , the choice of sensors, controllers and software is intricate. It is fairly easy to create an ideal model of a quadcopter
doi:10.3390/s151229785 pmid:26633410 pmcid:PMC4721706 fatcat:72tcgljxrzbzdeiviv5q5vgcz4