Pennsylvania State University. Unmanned Aerial Systems AUVSI SUAS Technical Report

Transcription

1 Pennsylvania State University Unmanned Aerial Systems 2018 AUVSI SUAS Technical Report Abstract This technical paper details the systems engineering process conducted by Penn State Unmanned Aerial Systems (PSU UAS) in its efforts towards competing in the AUVSI SUAS 2018 competition. The interdisciplinary team, comprised of undergraduate students studying Mechanical Engineering, Aerospace Engineering, Electrical Engineering, and Computer Science, worked towards its goal of constructing a custom designed fixed-wing autonomous aerial vehicle and communications system capable of completing the AUVSI SUAS mission tasks. Severe restrictions placed on the team s ability to conduct flights of university-owned property resulted in the delayed ability to implement the custom designed model for competition use. Anticipating this, the team co-developed a system utilizing an off-the-shelf model plane as a back up. Adjusting to the circumstances presented to the team and persevering to solve the problem, Penn State UAS plans to compete in the 2018 SUAS competition using the custom software developed by team members over the past year installed on a purchased airframe.

2 Contents 1 Systems Engineering Approach Mission Requirement Analysis Design Rationale System Design System System Aircraft Autopilot Obstacle Avoidance Imaging System Object Detection, Classification, and Localization Communications Air Delivery Cyber Security Safety, Risks, and Mitigations Mission Risks and Mitigations Developmental Risks and Mitigations Penn State Unmanned Aerial Systems

3 1 Systems Engineering Approach 1.1 Mission Requirement Analysis In order to prepare for the 2018 AUVSI SUAS competition, Penn State UAS first undertook an analysis of the assigned tasks, taking into consideration the weighting of each task, as well as the respective cost of pursuing each of them, in terms of their burden on the team s time and resource limitations. Certain tasks serve as basic minimum requirements to be able to functionally and safely compete, and therefore are given priority regardless of their corresponding points. Examples of such minimum requirements are mission testing, aircraft fail-safe modes, ability to relay telemetry to judges, and capability of safe flight, take-off and landing. Some points can be earned with negligible burden on team resources, such as writing of the technical design paper, operational excellence, as well as timeline, and therefore need not be sacrificed to conserve resources. Figure 1: Scoring weight breakdown With respect to the resource and timecostly tasks in the Mission Demonstration, those with a higher impact on the competition score were deemed to be a greater priority for time and resource allocation. Autonomous Flight accounts for the greatest portion of points in this category, therefore Autonomy and Waypoint capture were prioritized, followed by Obstacle Avoidance and Detection. This leaves Air Delivery as the most readily sacrificed task, if team resources were to become depleted to point that not all tasks could be fully attempted. Figure 1 illustrates the spread of points available in the competition, discretized into categories to facilitate the Mission Requirement Analysis. Once an understanding of the priority of tasks had been reached, the team proceeded towards developing a system to meet these requirements. 1.2 Design Rationale In approaching the design of any aspect of the craft, there exist certain general needs/limitations that must be accounted for, regardless of the specific task being considered. The most important of these needs were determined to be safety of the design, budget limitations, task complexity/team experience, and competition points. These were divided into four separate Design Layers (Figure 2), in order of their precedence. Penn State Unmanned Aerial Systems 1

4 Figure 2: Design Layers Safety: Risk to persons or property must be considered in every design decision, and all precautions taken in order to minimize it. Safety is considered to be the most important of the design layers Budget: As with any project, financial constraints must be taken into consideration for purchasing equipment and materials to execute the design. Expenses should ideally provide long-term benefit for the team and never leave the organization without emergency funds. Complexity and Team Experience: Designs should be sufficiently simple, modular, and achievable given team experience and time resources of team individuals. Competition Points: As many competition points should be achieved as possible.the system should be designed with mission tasks in mind with the goal of obtaining points through competition performance. Given that Penn State UAS has never flown in a competitive environment before, design and construction of such a system presents itself as uncharted territory for many team members. As such, the team made the decision to begin work on internal components for the system using an off-the-shelf model for testing, while simultaneously designing a custom craft for the airframe, with the hope that the custom model could be built (and properly tested for full functionality) for use in the 2018 competition. Should the team not feel 100% confident in the capabilities of the custom design, the off-the-shelf-model would be used for the 2018 competition, while continuing to develop the custom model for long-term use in future competitions. In addition to the influence of team experience on the design process, a major factor affecting the team s approach to the testing aspect of design was the introduction of new UAV flight policies by the Pennsylvania State University. PSU Risk Management policies forbid all flight of team-owned vehicles, which resulted in the team being unable to fly the system until late April of 2018 after undertaking a nearly two year long process of risk mitigation. In response to these severe delays on system testing and development, the team was forced to drop some tasks from the mission that had original been aimed for, including abandoning the airdrop task, as well as relying on manual object classification. 2 System Design 2.1 System 0 The preliminary system design was designated as System 0. With respect to the Autonomy category, this initial design aimed to achieve autonomous waypoint navigation with manual take-off and landing. Full autonomy would be aimed for in subsequent designs, once partial autonomous navigation is achieved. For the Obstacle Avoidance category, System 0 aimed to upload telemetry to the Interoperability System at 1 Hz. Regarding Object Detection, System 0 focused on GPS tagging and orientation of imagery. Penn State Unmanned Aerial Systems 2

5 Given a defined set of objectives for System 0, the team proceeded to construct a model with the corresponding hardware to obtain those objectives, as described in Figure 3. In compliance with the design layers previously developed, safety is of the utmost importance, and therefore structural components of the fuselage, wings and empennage, as well as some electronic components such as motors and batteries are given priority in the design process, to ensure that the system can fly safely prior to shifting focus towards earning points through autonomy and image capture. System 0 builds upon an off-the-shelf model for its structure. The team opted to use the Skywalker EVE 2000 for budgetary reasons, as this model was already owned and functional. The EVE served as a viable option for use given its ready-to-fly condition, as well as ample cargo space for the internal system. In terms of flight control, the design decided to use a Pixhawk 2.0, as several team members have prior experience with this brand of flight controller. For image capture, the team opted to utilize the Flea 3 HD camera, provided at a discount by a team sponsor. In order to power the system, a combination of Lithium Polymer (LiPo) and Lithium Ion (Li-ion) batteries were opted for to increase efficiency. Units requiring elevated discharge rates will uti- Figure 3: System 0 Architecture lize LiPos, while those requiring lower rates will draw from the lighter, more energy dense Li-ions. This customized battery combination lessens the weight of the fuselage, which can enable longer flight times. System 0 planned to use a Jetson as the on-board computer, as this was already owned and would not introduce any additional costs. The system will communicate with the ground station Lenovo computer using the telemetry radio system, to communicate data to competition judges and allow for monitoring of the system by the team. 2.2 System 1 Using System 0 as a building block, the team proceeded into the second phase of development, designated as System 1, intended to be used for competition. This system aimed to improve upon the details of the System 0 zero design, creating a more sophisticated system with additional capabilities. In this section, a brief overview of the system will be provided. Subsequent sections will detail subsystems of this design. Figure 4 illustrates the system architecture of this secondary design. Penn State Unmanned Aerial Systems 3

6 Figure 4: System 1 Architecture There exist some notable additions to the system in comparison to the System 0 design. Components essential to the mechanical functionality of the vehicle, such as the servos, motors, and batteries remain in the design for obvious reasons, while some more sophisticated features have been introduced. On board, such new design features include gimbal controller, Passive POE Injector, and Wifi radio. In addition, substantial developments have been made at the ground station, now including a Wifi radio, router, as well as manual classifiers and classification server. The following sections of this design paper will detail the major subsystems of the design. 3 Aircraft While significant progress was made on the design of the custom airframe, inability to conduct test flights until late in the competition season rendered it infeasible to confidently enter the custom model into the competition phase. As such, the team opted to install the system within a Skywalker 1880, an Expanded PolyOlefin (EPO) fixed wing plane. The Skywalker features Carbon Fiber tail boom, and measures out to an 1880 mm wing span, and 1180 mm length. This model was chosen due to its good payload capabilities, highly rated build quality, proven capabilities with hand launch, as well as team member prior experience with flight of the Skywalker. Batteries were chosen in order to minimize weight while maintaining proper power for the system. Increases in weight mean increased energy expenditure on actuations of gimbals/servos, as well as the motor Penn State Unmanned Aerial Systems 4

7 to propel the plane forward. Using a high energy density battery also reduces weight, further minimizing power consumption. In order to achieve the desired 40 minute flight time in the Skywalker 1880 airframe with a 600 gram battery and a 500 gram payload, we required 89 Watt-hours for flight energy. This was determined through flight tests conducted the previous year using a 500 gram dummy weight and a 6000 mah LiPo battery. In addition we required 10 Watt-hours to power the flight electronics (onboard computer, camera, wifi radios, gimbal). The desired battery to meet the system needs would have a minimum of 99 Watt-hours, weighing under 600 grams. As such, it was desired to use a battery energy density of 165 Watt hours/ Kg. To choose the battery most optimal for use in the system, research was conducted on a variety of models [1][2][3]. Model: Turnigy 4s Gens Ace 4s Sanyo 18650GA Capacity [mah]: Energy Density [W-hr/kg]: At first glance, it would appear that the Sanyo battery yields the highest energy density, and thus would be most desirable. However, the spec is a measurement of only the individual cells, and multiple cells are needed, thus the spec is higher than the actual result would be in flight. The team decided to move forward with use of the Gens Ace 4s battery, which meets the capacity and energy density requirements. Initially the team had considered use of lithium-ion batteries which allow for high degrees of customizability, however require significant design to obtain an efficient battery pack, as well as access to spot-welding facilities. As such, it was deemed that Lithium-Polymer batteries would be preferable. The chosen model reached 80% of the performance of the Lithium-ion batteries, while requiring significant less time to design a pack. 3.1 Autopilot The autopilot system will operate using a Pixhawk PX4 flight controller. The decision to utilize this flight controller was based on team member familiarity with Pixhawk technology, controllability, as well compatibility with MAVLink. As seen in Figure 4, many different sensors such as the GPS module and airspeed sensor are connected to the Pixhawk. The attached sensors (Figure 5) are essential for accurate operation of the flight controller, and as such, Software and Aeromechanical teams must collaborate to ensure proper functionality of these features. The team created a custom program to interact with the Pixhawk using a telemetry radio and MAVLink from the ground. The program automatically requests the current GPS coordinates of the aircraft multiple times per second and then submits a post request directly to the interoperability server. The Pixhawk was also chosen because it can sustain multiple MAVLink connections simultaneously. This allows the team to have separate programs to transmit waypoints for the aircraft to follow, receive GPS coordinates in accordance to the competition rules, and geotag images separately. The Pixhawk also has customizable predefined safety measures which mitigate risk in the event of loss of connection. Lastly the Pixhawk implements high level abstractions such as waypoints automatically. This allows the autopilot team to focus more on the path of the aircraft as opposed to the specific motors that allow it to change course. The Pixhawk will connect to a TM, as well as the on-board Jetson computer, which will pull telemetry data from the flight controller when needed. Note that this telemetry data is not sent to the Judge s Server (JS) right away. Instead, it is used for images labeling so that later on to successfully identify the location of each target. Further details of communication within the system will be enumerated in section 3.5, Communications. Penn State Unmanned Aerial Systems 5

8 Figure 5: The Pixhawk communicates with the Jetson and onboard systems 3.2 Obstacle Avoidance It was initially intended for the aircraft to use a customized implementation of the A* pathfinding algorithm to avoid obstacles in real time. Theoretically the system would take in obstacles from the interoperability server and using a grid approximation of the competition boundary, construct an optimal path around the obstacles while completing the other competition tasks. However, this system was not able to be implemented due to inability to conduct test flights throughout the semester, impeding the ability to create and iterate on such a system. As a result, real time obstacle avoidance was not fully implemented for the competition. Instead, the aircraft flies using manually predetermined waypoints, being sure to never come within a determined safe distance of the competition boundaries. 3.3 Imaging System The camera is one of the most important components of the system, as the main goal of the competition is to autonomously capture high quality images of the ground for further processing. As stated in the rationale for the System 0 design, the team opted to utilize a Flea 3 HD camera for the imaging system. This decision was made given that the camera had been significantly discounted to the team, introducing small additional cost, as well as served the necessary functionality. The Point Grey Flea3 camera with Tamron lens captures images at a resolution of 4096 x 2160 or 8.8 MP. The high resolution of the imaging camera allows for greater detail when determining target shape and color. The db dynamic range as well as the relatively large 1.55 µm were positive contributing factors due to the need to reliably identify color. Capturing 20 frames per second, the goal of the imaging software is to set up all the parameters in such a way that would allow us to capture high quality images at a very high altitude. 3.4 Object Detection, Classification, and Localization USB connection between the onboard Jetson computer and the camera allows for reading/writing of data from the camera buffer. Imaging processes will take place in two different modules (Figure 5), split between the on-board Jetson computer and the ground station Lenovo laptop. The Jetson computer will be responsible for image capture and transfer, while classification and control take place on the Lenovo. Penn State Unmanned Aerial Systems 6

9 GPS capture code and Image Capture code will be launched on separate threads from the main in order to ensure that one does not block another. The use of shared variables will allow the team to receive the most recent data any time a new image is captured. Images will be saved on the Jetson s SD card, at which point the Communications executable will be responsible for transfer of the image to the system. These executables are illustrated in Figure 6. The Ground Station (GS) Lenovo computer will handle image classification and transfer that data to the Judge s server (JS). (a) Imaging Systems Architecture Figure 6: Imaging System Design (b) Image Exec. The Classification & Control application provides a user friendly interface to allow for manual shape identification. This application consists of two modules: Shapes Classification and UAV Control. The Classification module will provide the user with the ability to get the images from the ready queue, select the regions with shapes using the cursor and then label each target with the required properties such as shape, color, alphanumeric value, and orientation. Finally, the user should be able to send all the information to the JS with an API call. Note that each image that has to be classified will be labeled with the GPS coordinates. The classify application will have to calculate the exact GPS coordinates, which is embedded into the images and transferred from the Jetson for each selected target. The GS will push data back to the JS for telemetry updates and classified targets submission. 3.5 Communications A functional communications system is essential to the performance of the vehicle. Without proper communication, the components of the aircraft cannot interact in the necessary manners to ensure completion of the designated tasks. There are three major components in our systems that will be running all of our code. The first is the Jetson which is responsible for onboard operations such as image capture, GPS matching, and file transfer. The second is the ground station computer which is responsible for ground operations such as communicating with the UAV, communicating with the JS, and providing protocol for manual images detection. The third and final component is the processing server, the supporting unit of the ground station where the majority of computational power is consumed. Each of these components relies on proper communication within the system. An overview of the software communication Penn State Unmanned Aerial Systems 7

10 architecture can be found in Figure 7. Figure 7: Software Communication System Architecture The aircraft and the Ground Station (GS) will communicate over two separate radio links: a WiFi link (WiFi) and a Telemetry Radio (TR). The WiFi is designed to perform the initial setup of the system (i.e. starting up all the components on the Jetson through an SSH connection) and to transmit images from the aircraft to the ground and will be connected to the Jetson via antenna. The TR is responsible for the MAVLink communication directly with the flight controller (Pixhawk). The GS will act as a server and it will provide some API for the Processing Server and for the support stations. Different API calls will perform different operations acting as separate executables. Also, there will be another executable (Classify & Control) that would enable the users to manually classify the images and send MAVLink commands to the UAV for obstacle avoidance. A simplified version of this software will be run and the support devices, as only the main machine will be sending MAVLink commands to the flight controller. The sole purpose of the Jetson computer that is on the plane is to capture the image with the camera, label it with the GPS coordinates from the Pixhawk and send them to the ground. Based on our experience, image processing requires many complex computations and therefore takes a lot of time and consumes a lot of power. For that reason, the decision was made to stream all the images to the ground and do processing at the ground station. In order to develop a system that is as modular as possible such that different parts of the system do not conflict with one another, the process is broken into two executables. The Comm Exec is responsible for image transfer, while the Img Exec is responsible for capturing images and labeling with GPS coordinates. Penn State Unmanned Aerial Systems 8

11 3.6 Air Delivery Due to delays in the development of the system resulting from university policy restrictions placed on flight testing, the team has opted to focus on completion of some of the more heavily weighted tasks. After considering the time resources remaining, the decision was made to not pursue the Air Delivery task. 3.7 Cyber Security Maintaining and ensuring secure communication between all system components is critical to the success of the mission. There are two main communication links that must be secured: 1. UAV and Ground Station The UAV and the ground station communicate through a secured WPA2 network. This ensures that only authenticated devices are introduced into the network. Furthermore, the telemetry data is transferred over a dedicated radio antenna from the UAV to another dedicated antenna connected to the ground station. This connection is isolated from the remaining network to guarantee stability and security. 2. Ground Station and Interoperability Server The ground station is connected to the AUVSI interoperability server through an ethernet line. This connection avoids interference that could occur from a wireless connection, and provides another layer of security. Connecting to the AUVSI interoperability server also provides the ground station with an authentication cookie, which uniquely identifies the ground station. Since the ground station is not connected to any other networks besides the AUVSI server and the WPA2 secured network, the risk of cyber attacks is reduced. 4 Safety, Risks, and Mitigations The importance of safety during the development and operation of the system cannot be understated. As such, numerous safety and training standards have been implemented at all levels and areas of development and testing. 4.1 Mission Risks and Mitigations Multiple levels of contingencies have been installed in the flight system in case of failure. A manual pilot mode is available in case of a malfunction with the system autopilot. If the connection to the flight system is lost during operation, the aircraft will execute a return to home command. Another potential issue would be loss of WiFi connection. This is addressed by transmitting telemetry data to the Ground station via TR. However processing would not be able to occur while the plane is in flight. To ensure that all data is not lost, an SD on the Jetson will save each image for processing after landing, bypassing the computer vision algorithm. 4.2 Developmental Risks and Mitigations Development of an aircraft system introduces a variety of risks. To reduce the likelihood of injury to members during building of the airframe, all machining and fabrication is conducted by team members that hold university-issued certifications at the Bernard M. Gordon Learning Factory. Penn State Unmanned Aerial Systems 9

12 Entering the academic year, the Risk Management Department of the Pennsylvania State University communicated to the team that the department considered flight of university-owned property to be high risk, and subsequently forbid the team from flying. In response, the team pursued a variety of risk mitigation strategies to appeal to the Risk Management department for permission to fly. In order to reduce the risk of injury to persons or property during flight, team pilots completed the rigorous FAA 107 pilot certification course to obtain the proper clearance for operating an aircraft under university Risk Management Policies. The team routinely met with officials at Risk Management to discuss precautionary measures, develop safety plans, and met with many professors and researches experienced in UAV flight and research. The team obtained flight locations off of campus that are specifically designated for flight testing. The team also will obtain personal insurance in addition to the insurance provided to all university organizations. After over a year of appeals, the team anticipates being able to resume flight testing in early May, where members will work to ensure that the system is ready for competition. Penn State Unmanned Aerial Systems 10

13 References [1] Turnigy 5000mAh 4S 25C Lipo Pack w/xt-90 [2] Gens Ace 6750mAh 14.8V 45C 4S1P Lipo Battery Pack with XT90 plug GA-B-45C S1P-XT90 [3] Panasonic-ncr18650-ga-spec-sheet, cdn.shopify Penn State Unmanned Aerial Systems 11