Uav Performing Autonomous Landing on Usv Utilizing the Robot Operating System

This paper outlines the design and implementation of the launching and landing of an unmanned aerial vehicle (UAV) onto an unmanned surface vehicle (USV). The US Navy and Northrop Grumman’s X-47B drone have just recently achieved the first autonomous landing of an aircraft onto an aircraft carrier, but it came at the price tag of over one billion U.S. dollars. The authors of this paper will create a scaled-down model of this system that will cost in the order of thousands of dollars. The aerial vehicle utilized in this study is a quadrotor helicopter, which greatly simplifies the launching and landing process. The surface vehicle used is an autonomous boat that features a catamaran-style hull. Both vehicles will use the Robotic Operation System (ROS) as a framework for their software. A successful mission for this system would be one in which the quadrotor launches from the launch pad of the boat, navigates to a series of GPS waypoints, lands at a predetermined location to pick up an object, navigates back to the boat, and then lands successfully to the USV. INTRODUCTION Since World War II, the ability for an aircraft carrier to deploy and recover aerial vehicles thousands of miles from the closest air base has proven to be a great tactical advantage to those forces capable of utilizing them. Despite the advantage that they provide, the success of carriers is largely dependent on the skill and experience of the pilots flying the aircraft. The use of UAVs makes it possible to reduce failure due to human error, while also preventing the loss of human lives in the process. Landing a rotary wing aircraft onto an aircraft carrier has several challenges associated with it. Unlike descent over land, the aircraft pilot has to compensate for the pitching of the ship to make sure the boat doesn’t accidentally slam into the landing gear. This could damage the aircraft as well as cause it to become unstable. As the aircraft approaches the surface of the ship, it becomes susceptible to the ground effect which may also cause it to become difficult to control. A number of researchers have investigated methods to autonomously land UAVs using vision guidance. The simplest method, shown in [1], uses a combination of GPS waypoints and vision to get the UAV in the vicinity of a landing pad. The UAV then uses vision to search for the location of the landing pad. Once the UAV has located the pad using an attached camera, the UAV then begins its descent. In [2,3], more sophisticated methods were developed for UAVs to use vision to identify hazardous features to determine whether or not a surface was safe to land. Also pertinent to this study is the work already done on using UAVs in conjunction with unmanned ground vehicles (UGVs) to accomplish a mission. The research shown in [4] uses a manually flown AUV and an autonomous UGV. The AUV shared the information gathered by its camera to assist the UGV with path planning and obstacle avoidance. A group in Switzerland has also done extensive work in using UAVs magnetically mounted onto the ceiling to help a swarm of UGVs navigate between two targets in a foraging task system that closely mimics ant colony movement [5]. Other focuses of the research show how eye-bots assist foot-bots in traversing a gap clearing environment [6] and using a single eye-bot to assist foot-bots in traveling over a hill in terms of angle steepness and calculating group formation in order to traverse an angle too steep for a single vehicle to climb [7]. Wenzel [8] was able to use vision to autonomously take off and land a quadrotor onto a small ground vehicle with a 90% success rate. However, since their vision system relies on an infrared (IR) camera, their system only works in indoor environments with controlled lighting. Additionally, by landing on a land vehicle in an indoor environment, one can make the assumption that the orientation of the landing surface is constant. Landing on an USV presents an additional challenge, ASME District F ECTC 2013 Proceedings Vol. 12 119 since the orientation of the vehicle’s landing surface is also free to roll and pitch. In this paper, the design and implementation of a heterogeneous robotic system composed of a UAV and an USV will be detailed. ROS is used to handle high-level behaviors, perform controls, and facilitate communications between the USV and UAV. Using vision, the UAV will be able to safely land on both a stationary target as well as a landing pad mounted on top of the USV in both indoor and outdoor lighting conditions. UNMANNED SURFACE VEHICLE DESIGN The unmanned surface vehicle used in this study was designed and built by undergraduate and graduate students at the Machine Intelligence Laboratory (MIL) at the University of Florida. It was initially designed for the purpose of competing in the Association for Unmanned Vehicles Systems (AUVSI) and Office of Naval Research’s (ONR) 6th Annual International RoboBoat Competition. The USV earned first place and has since been devoted to robotics research. Figure 1. CAD rendering of the USV and UAV system The USV used in this study features a catamaran-style hull to provide stability as well as a shallow draft. The two pontoons are held together with a frame composed of 8020 aluminum extrusions. For propulsion, the USV has four trolling motors. The front pair is pointed inwards at an angle of 30° while the back pair is pointed 30° outwards. This allows the USV to translate in any direction as well as rotate in place. Combined, the four trolling motors are capable of providing up to 22 lbs of forward thrust. It has a water displacement of 115 lbs and overall dimensions of 72” x 30” x 36” (length x width x height) [9]. A complete computer-aided design (CAD) model of the USV and UAV system was created using SolidWorks and can be seen in Figure 1. There are several benefits to creating a CAD model for a vehicle. First, the model allows the designer to visualize how components fit together. Second, it facilitates the use of engineering analysis tools. For example, airflow simulations conducted on the model helped determine the geometry of the vents used for the USV’s cooling system. Third, the individual components that make up the model can easily be fabricated using the lab’s Dimension SST 1200es 3D printer or Roland MDX-540 SRP 3D milling machine as shown in Figure 2. Figure 2. Fabricating parts using the 3D milling machine UNMANNED AERIAL VEHICLE DESIGN The UAV used in this study is a quadrotor helicopter with a commercially available DJI Flame Wheel F450 frame. Off-theshelf parts were selected for the UAV to minimize the cost of the overall system, as well as to promote open-source collaboration. The ABS landing gear on the UAV is designed so that in the event the UAV crashes over land, the landing gear will absorb most of the impact and break off. This will leave the rest of the frame with minimum damage. Figure 3 shows the UAV used in this study. Figure 3. UAV used in this study All of the electrical components on the UAV are powered by a single 6000 mAh 3-cell lithium polymer (LiPo) battery. With a full charge, the UAV has an approximate flight time of 15 minutes. The weight of the UAV is 3.8 lbs, and it is lifted by four brushless DC motors, which can each provide 1.2 lbs of thrust. ASME District F ECTC 2013 Proceedings Vol. 12 120 Basic stabilization and navigation is handled with an ArduPilot Mega (APM) 2.5+ autopilot control board. Control and sensor processing abilities are provided to the UAV via a quad-core ARM ODROID-U2 installed with Linux-Ubuntu and ROS. The ODROID can retrieve information such as current orientation and GPS location from the APM. The ODROID is also capable of sending the APM information such as a desired attitude or GPS waypoint. The main sensor of the UAV is a Logitech HD Pro Webcam C920 camera which is used for computer vision as well as obstacle and target identification. Communication between the two vehicles is handled by XBee RF devices which have a range of approximately of 300 feet. ROS ENVIRONMENT Both the USV and the UAV operate within a ROS environment. ROS was chosen due to the open source nature of the software. This allows for various sources of code to be used or for code developed in this research to be easily shared. ROS also allows an easy method for all code, in the form of nodes, to be duplicated and moved to multiple vehicles. This is used throughout the various levels in the heterogeneous cooperative system. Some of the groundwork code for the UAV was completed in previous studies [10,11]. Since all nodes created are started with and must have access to a ROS Master Core, a loss of connection to the ROS Master would result in all nodes and controls shutting down, effectively stopping the vehicle from functioning. Each vehicle is therefore set up as a ROS Master to avoid this failure. Though each vehicle is independent and is the master of its own instance of ROS, vehicles are able to communicate via RF so that they can cooperatively plan their desired tasks to complete the mission. Figure 4. ROS_Bridge communication between vehicles Communication, control, and data are all handled within the ROS environment through multiple nodes. The ROS_Bridge node runs separately on each ROS Master, handling communication between each vehicle within the overall ROS environment (Figure 4). ROS_Bridge is configured using specific parameters which define desired topics and services to be shared between multiple ROS Masters. Once a topic or service is modified with new data within a ROS Master, ROS_Bridge acknowledges the change, and broadcasts the resulting data via RF communications. Other RF devices then receive this data, and subsequently update their own ROS Masters. Within this research, ROS_Bridge primarily handles mission objects in the form of desired GPS points as well as mission status information. Direct UAV control and stabilization is left to the APM autopilot. However, mission navigation, as well as direct control for visual servoing, is handled via the APM_Driver node. The APM allows external devices to communicate with the onboard autopilot software via a standardized Micro Air Vehicle Communication Protocol called MAVLink. Specifically, the APM_Driver uses MAVLink to receive attitude and status information from the APM which is then published to ROS. Other nodes may specify mission objectives in the form of GPS waypoints which are sent to the APM for waypoint navigation. Commands may also be sent that result in the APM arming, disarming, launching, landing, or modifying stabilization modes. During visual servoing, the APM_Driver node is used to send roll, pitch, yaw, and altitude adjustments to the controller. Figure 5. Data flow with Agent_Control ROS