Unmanned Aircraft Systems Sense and Avoid Avionics Utilizing ADS-B Transceiver

Transcription

1 AIAA Conference <br>and<br>aiaa Unmanned...Unlimited Conference 6-9 April 2009, Seattle, Washington AIAA Unmanned Aircraft Systems Sense and Avoid Avionics Utilizing ADS-B Transceiver Florent Martel 1, Richard R. Schultz 2, William H. Semke 3, Ziming Wang 4 and Mariusz Czarnomski 5 Unmanned Aircraft Systems Engineering Laboratory, University of North Dakota, Grand Forks, ND 58202, USA To integrate Unmanned Aircraft Systems (UAS) into the U.S. National Airspace (NAS), the Federal Aviation Administration (FAA) requires an equivalent level of safety, comparable to see-and-avoid requirements for manned aircraft. A proposed FAA rulemaking would mandate Automatic Dependent Surveillance Broadcast (ADS-B) out equipage by 2020, forcing aircraft flying in the NAS to broadcast their state vector to Air Traffic Control (ATC) and aircraft equipped with ADS-B in capability. With the advent of lightweight, lowpower, and low-cost ADS-B units, the equipage and the use of ADS-B transceivers on small UAS are no longer limited by payload and power capabilities, and represent promising opportunities for the regulated operation of small UAS in the NAS. The University of North Dakota (UND) has been actively developing enabling sense and avoid (SAA) technologies under a series of projects funded by the Department of Defense, and is now turning to ADS-B as the central component for mid-air collision avoidance. This paper presents the results of modeling collision avoidance algorithms using ADS-B derived information in a software-in-the-loop (SWIL) environment. Upon extensive testing, the system is designed to be integrated seamlessly into a hardware-in-the-loop (HWIL) environment, and ultimately in a flight test environment aboard a small UAS. I. Introduction nmanned Aircraft Systems (UAS) applications are proliferating among the military, government, and private Usectors due to significant improvements in capabilities and performance. However, the potential of UAS is constricted by many regulations, and the ability to fly unfettered in the U.S. National Airspace System (NAS) is conditioned by the elaboration of specific standards and regulations tailored for UAS. Following intense pressure from industry to see the Federal Aviation Administration (FAA) elaborate clear guidelines, the recently introduced FAA Reauthorization Act of 2009 [1] calls for the FAA to provide, no later than nine months after the date of enactment, a comprehensive plan with detailed recommendations to be followed by a proposed rulemaking to safely integrate UAS in the NAS by no later than September 30, This window of opportunity calls for the development of state-of-the-art collision avoidance algorithms and their implementation on avionics interfaced with the proper set of sensors. The Automatic Dependent Surveillance Broadcast (ADS-B) transceiver is a key step in creating the Next Generation (NextGen) Air Traffic Control (ATC) system. Furthermore, with the proposed FAA rulemaking that would mandate ADS-B out equipage by 2020 [2], ADS-B will most likely become an essential piece of a sense and avoid (SAA) system for UAS operating in the NAS. In addition, with ADS-B being a datalink, further capabilities may be provided such as Primary Surveillance Radar (PSR) derived information through the transmission of Traffic information Service Broadcast (TIS-B) messages from Ground Based Transceivers (GBTs). This implementation would provide an alternative to the addition of non-cooperative sensors onboard small UAS, which is not yet technologically feasible. Another interesting capability emerges from the work of MITRE on an ADS-B to Link16 ground-based gateway [3], which would provide civilian/military aircraft connectivity without the need to retrofit any Link16 equipped military aircraft. 1 Electrical Engineering Graduate Student, UND Dept. of Electrical Engineering, Grand Forks, ND Professor of Electrical Engineering, UND Dept. of Electrical Engineering, Grand Forks, ND Assoc. Prof. of Mechanical Engineering, UND Dept. of Mechanical Engineering, Grand Forks, ND Electrical Engineering Graduate Student, UND Dept. of Electrical Engineering, Grand Forks, ND Electrical Engineering Undergrad. Student, UND Dept. of Electrical Engineering, Grand Forks, ND Copyright 2009 by the, Inc. All rights reserved.

2 II. SAA System Architecture Figure 1 shows the embedded sense and avoid (SAA) architecture that our laboratory has implemented on a small UAS. The architecture is comprised of three separate systems. An ADS-B transceiver receives aircraft information from a cooperative source, which may be another ADS-B equipped aircraft or a GBT. The ADS-B transceiver is connected to an intelligence system, which is a flight computer intended for real-time processing. The intelligence system answers to a set of hierarchical rules to perform conflict detection. Upon determination that another aircraft or an obstacle represents a threat, collision avoidance is performed by issuing high-level guidance commands to the onboard autopilot that are intended to reroute the UAS on a safe path. Guidance commands are issued only if the aircraft is following a trajectory presenting a danger, and the computed maneuvers are consistent with FAA right-of-way rules. If a guidance command is issued by the intelligence system, the autopilot relinquishes direct control of the UAS over its intended trajectory and instead interprets the high-level guidance commands into actuator control signals to maneuver the UAS. Figure 1. Sense and avoid system architecture. Figure 2 shows the software-in-the loop (SWIL) test-bed implemented in MATLAB, which is derived from the SAA architecture shown in Figure 1. It is comprised of four elements that define not only functional equivalents of the actual hardware systems, but also the dynamics of the UAS and its environment. The SWIL test-bed is designed to be modular and flexible so that actual hardware can be swapped for hardware-in-the-loop (HWIL) simulation purposes or for actual flights. Figure 2. SWIL test-bed. 2

3 A. Simulation Engine The Simulation Engine block simulates the ownship aircraft and multiple intruder aircraft in an operational flight environment including terrain, elevated obstacles (e.g. communication towers, buildings, etc.), and airspace boundaries defined in a database. The setup has the ability to simulate as many intruders as would be received by an actual ADS-B transceiver, although functionally the number of possible simulated intruders is not limited. The Simulation Engine includes a flight dynamics model to simulate the flight of the ownship and intruders. The dynamic characteristics are modeled for a wide variety of aircraft, although the ownship model is defined for a particular small UAS. The Simulation Engine also includes a flight control loop to complete the modeling of the ownship aircraft. The Simulation Engine is fed the actuator control commands from the Autopilot Engine, and it outputs aircraft information including the position of both the ownship and the intruders. B. ADS-B Engine The ADS-B Engine block is used to simulate the operation of an ADS-B transceiver in its environment. It is fed aircraft information from both the ownship and the intruders and outputs the information in the same format and sampling rates as the actual ADS-B hardware. Due to the nature of GPS measurements and ADS-B transmissions, this engine models range, accuracy, and dropout limitations inherent to an ADS-B transceiver. C. Collision Detection/Avoidance Engine The algorithms contained in the Collision Detection/Avoidance Engine block are the same algorithms that will run on the flight computer onboard the aircraft during flight tests. This block contains a collision detection logic and a collision avoidance logic, as well as a database of terrain, elevated obstacles, and airspace boundaries. The collision detection logic estimates whether or not the ownship aircraft is on a collision course with respect to other aircraft or fixed obstacles. This algorithm tracks the position of air traffic based on the GPS positions obtained from the ADS-B Engine, even during temporary dropouts, and is defined to accept inputs from additional sensors that would contribute to estimating the position of air traffic. Filtering and sensor fusion techniques for position estimation are well known, and a detailed explanation of their design and operation is not necessary here. Instead, this paper focuses on the collision avoidance logic that is explained in detail in Section 3. The collision avoidance logic is activated if threats are detected. A threat is characterized as an aircraft or a fixed obstacle that is predicted to enter a collision or near collision course with the ownship aircraft within a certain time frame. The collision avoidance logic issues, if necessary, an avoidance command in the form of high-level commands such as heading, speed, and altitude changes in order for the ownship to steer clear of threats. The computational complexity allows the system to run in real-time in the SWIL environment, although the goal is to minimize complexity to lighten the load on the simulation computer and to allow the system to run at a much faster rate than real-time in order to test the algorithms in a large number of encounter scenarios in a short period of time. Since ADS-B data updates once per second, real-time operation requires all processing to take place during that time interval. D. Autopilot Engine The Autopilot Engine block handles the conversion of high-level commands into actuator control signals. The control signals are fed to the Simulation Engine block to perform the ownship aircraft maneuvers. In the absence of input from the Collision Detection/Avoidance Engine block, the Autopilot block outputs control signals that handle the navigation of the ownship aircraft along a set of waypoints according to a specific waypoint guidance control logic, as would be performed during the regular operation of an autopilot. E. Transition to HWIL Simulation Computer simulation represents the core of this SAA system study. The goal is to model effective collision avoidance procedures while conducting a risk analysis to minimize false alarm and false dismissal rates in a realistic simulation environment. Upon extensive testing and validation of the collision avoidance algorithms by performing a multitude of simulations in the SWIL test-bed, the system will progressively be migrated to HWIL simulation, with the concurrent development of device drivers to render the system compatible with a variety of commercialoff-the-shelf (COTS) sensors and autopilots. In this context, the ADS-B Engine and Autopilot Engine blocks will be replaced by the corresponding hardware. The model shown in Figure 2 has the ability to automatically generate realtime embeddable code so that the Collision Detection/Avoidance Engine block can be ported to a target flight computer seamlessly. 3

4 III. Autonomous Collision Avoidance Algorithms This section describes in detail the algorithms contained in the Collision Detection/Avoidance Engine block. The first iteration of the SAA system, designed to use ADS-B information, assumes that the GPS position of surrounding air traffic can be obtained, with inherent errors due to the nature of GPS measurements and of the ADS-B datalink transmission modeled in order to simulate the use of an ADS-B transceiver under operational conditions with high fidelity. Terrain, elevated obstacle, and airspace boundary information is obtained from a database preloaded before flight. The proposed algorithm has the aptitude to handle situations involving multiple simultaneous intruders. If one or more intruders are predicted to enter a collision or near collision course with the UAS according to its current flight path, the proposed collision avoidance algorithm will determine an updated flight path, in which any possible collisions will be tentatively avoided. The algorithm uses a behavior-based approach and is derived from research performed on unmanned surface vehicles [4]. The decision logic behind the collision avoidance algorithm that provides the high-level commands combines behavior results based on weighting to produce a single, optimal result. FAA right-of-way rules involving encounter situations can be directly coded into this set of behaviors. The first construct is to represent multi-objective optimization problems with piecewise linearly-defined objective functions over a finite decision space. The problems associated with an exponential growth of state space are countered by dealing with separate sub-situations independently. Sub-situations are behaviors which can occur simultaneously, but by producing independent objective functions for each behavior, a modular approach which grows linearly in complexity with each behavior can advantageously be used. Behaviors may be conceived to provide operational results such as reaching a target waypoint, avoiding fixed or moving obstacles, and following air traffic regulations. An extensive number of behaviors may be added to the collision avoidance algorithm if designed appropriately to enhance its capabilities. These behaviors may include following a trajectory, loitering over a target, maintaining a constant heading, speed, or altitude, and following the quickest or steadiest path. Behaviors may also be combined, depending on functional requirements. Several behaviors are explicitly described in this paper based on their complexity and to provide a more complete understanding of the underlying mechanism behind the proposed collision avoidance algorithm. Each behavior may or may not produce, according to thresholds, an objective function. Objective functions are defined over a decision space limited by a set of explicit constraint constructs representing intervals of values and harmonized over the same decision space in order for the behaviors to be combined as a weighted sum of the objective functions. The weights reflect priority, or the degree to which the decision is made as a trade-off between behaviors based on the overall context or the mission requirements. The resulting objective function is used to determine the preferred high-level command to be fed to the autopilot. A. Closest Point of Approach Many of the objective function evaluations are based on calculating the Closest Point of Approach (CPA), which refers to the situation in which two objects reach their closest possible distance. In the context of this research, the two objects are the ownship aircraft and a waypoint, or the ownship aircraft and an obstacle. Consider two objects which positions at time t defined in terms of latitude, longitude, and altitude given as os_lla(t) and cn_lla(t), and their velocity vectors per unit of time given as os_dxyz(t) and cn_xyz(t). The velocity vectors are defined with Cartesian coordinates in a North-East-Up (NEU) reference frame. Let the current time be t 0. The equations of motion for these two points are os_lat(t) = os_lat(t 0 ) + t os_dxyz(t) and cn_lat(t) = cn_lat(t 0 ) + t cn_dxyz(t). The distance between the two points is distance(t) = os_lat(t) - cn_lat(t) = distance(t 0 ) + t (os_dxyz(t) - cn_dxyz(t)). Since the distance between two points cannot be negative, distance(t) has a minimum when distance(t) 2 has a minimum, therefore distance(t) has a minimum when d( distance ( t dt CPA 2 ) ) 2 = 2t ( os _ dxyz cn _ dxyz) + 2distance ( t ) ( os _ dxyz cn _ dxyz) = 0. CPA The above equation gives the time to CPA as distance ( t ) ( os _ dxyz cn _ dxyz) os _ dxyz cn _ dxyz 0 t CPA =. (2) (1)

5 If os_dxyz(t) - cn_xyz(t) 0, the two points will remain approximately the same distance apart, so t CPA can be set to zero. If t CPA < 0, the two points are travelling away from each other, so t CPA can be set to zero. The CPA distance between the two points is then found by calculating distance(t CPA ) for a given ownship velocity vector value. The ownship velocity vector values correspond to defined heading, speed and altitude values, given the nominal climb and descent rate of the ownship aircraft. B. Reach Target Behavior In order for the ownship aircraft to navigate towards a target waypoint, a Reach Target behavior is introduced. In Figure 3, the Reach Target behavior is illustrated, with the ownship aircraft represented relative to a target waypoint. The position of the waypoint is defined in terms of latitude, longitude, and altitude. In this example, the output of the Reach Target behavior is an objective function defined over a decision space in two dimensions as a function of possible heading and speed commands for illustration purposes. However, an objective function is actually defined in terms of possible high-level commands to ensure complete spatial maneuverability, and in the present implementation the high-level commands refer to heading, speed, and altitude values. In Figure 3, the possible heading and speed values are respectively represented as the angle and radius of the toroidal objective function section. Minimum and maximum speeds are represented respectively by the inner and outer boundaries of the toroidal section and defined according to the performance of a specific small UAS. HIGH-LEVEL COMMAND OBJECTIVE FUNCTION Figure 3. Exemplary representation of the Reach Target behavior. The objective function generated by the Reach Target behavior is constructed to privilege high-level commands which maneuver the aircraft closer to the target waypoint, and is determined so as to assign higher arbitrary values to high-level commands which minimize the distance at CPA between the ownship aircraft and the target waypoint. A utility metric is applied according to the following sigmoid-shaped function ( tcpa q)/ r f RT = 1/(1 + p), (3) with p, q and r set to tune the behavior to the performance of a specific UAS. In the representation of the objective function in Figure 3, lighter areas represent higher arbitrary values and darker areas represent lower arbitrary values. A recommended high-level guidance command is given by the value for which the maximum of the objective function is attained. Using the high-level guidance command resulting from the Reach Target behavior is futile on its own, since the onboard autopilot would already fulfill the function of guiding the ownship aircraft towards a target waypoint without intervention. In order for this behavior to be meaningful and for the ownship aircraft to avoid collisions, the Reach Target behavior must be combined with other behaviors. C. Avoid Small Obstacle Behavior The goal of the Avoid Small Obstacle behavior is to prevent the ownship aircraft from coming dangerously close to a small obstacle, such as the situation shown in Figure 4. In this example, a Reach Target behavior and an Avoid Small Obstacle behavior are activated respectively by the target waypoint and the intruder aircraft, and each behavior generates a separate objective function. The objective function generated by the Avoid Small Obstacle behavior is constructed to privilege high-level commands which maneuver the ownship aircraft far enough from the obstacle, and is determined so as to assign higher arbitrary values to high-level commands which maintain at least a minimum distance at CPA between the ownship aircraft and the obstacle. The objective functions are combined in a weighted sum to produce a resulting objective function. In Figure 4, the darker area of the resulting objective function represents heading and speed values for which the ownship aircraft may potentially collide or nearly collide with the obstacle. 5

6 INTRUDER HIGH-LEVEL COMMAND OBJECTIVE FUNCTION Figure 4. Exemplary representation of the Avoid Small Obstacle behavior involving one intruder. The guidance method may advantageously process fixed and moving targets with the same Reach Target behavior, and may process fixed and moving small obstacles with the same Avoid Small Obstacles behavior. Additionally, the Reach Target behavior and the Avoid Small Obstacles behavior may use the same set of equations to generate their respective objective functions, as the construction of the objective functions is based on calculating the distance at CPA. Only one target at a time may be considered to produce the guidance commands, since the ownship aircraft cannot navigate in multiple directions for obvious reasons. Multiple small obstacles may be considered simultaneously by the guidance method, however. Another significant advantage when dealing with multiple small obstacles is that each small obstacle may result in the computation of only one additional objective function; therefore, computational complexity increases at most linearly with respect to the number of small obstacles considered for avoidance purposes. A situation involving multiple obstacles is shown in Figure 5, where the ownship aircraft intends to reach a target waypoint while avoiding two obstacles represented by aircraft in motion. Each obstacle may, according to the collision deconfliction threshold, activate the Avoid Small Obstacle behavior and generate a unique objective function. This example illustrates how multiple objective functions generated by the Avoid Small Obstacle behavior can be combined to produce the resulting objective function. INTRUDER INTRUDER OBJECTIVE FUNCTION HIGH-LEVEL COMMAND Figure 5. Exemplary representation of the Avoid Small Obstacle behavior involving two intruders. D. Avoid Large Obstacle Behavior The Avoid Small Obstacle behavior cannot be used to avoid large obstacles, defined as obstacles large enough to be inaccurately represented by a single point in space for avoidance purposes. Large obstacles include terrain, elevated obstacles, and airspace restrictions such as air corridors, airspace classes and no-fly zones. Large obstacles may activate an Avoid Large Obstacle behavior, which replaces the Reach Target behavior, since the Avoid Large Obstacle behavior will guide the ownship aircraft towards a target waypoint by finding the path of minimum distance between the aircraft and the target and generates an objective function dependent on the additional detour distance over the computed shortest path distance caused by maneuvering around large obstacles. The Reach Target 6

7 behavior may systematically be replaced by the Avoid Large Obstacle behavior, but the Reach Target behavior is privileged in the absence of large obstacles within a certain range or in the direct path of the ownship aircraft to reduce computational complexity. The Avoid Large Obstacle behavior is explained below in Figure 6 as a two- dimensional problem, but is used in three dimensions by the algorithm so as to enable the ownship aircraft to perform altitude changes in order to follow the path of minimum distance to the target in three dimensions. x x OWNSHIPAIRCRAFT (a) (b) SHORTEST PATH FUNCTION x SHORTEST PATH FUNCTION x SHORTEST PATH OBJECTIVE FUNCTION (c) Figure 6. Exemplary representation of the Avoid Large Obstacle behavior showing (a) a terrain database map containing elevated terrain directly in the path between the ownship aircraft and the target waypoint, (b) a binary map extracted from the terrain database and from an airspace boundary database displaying large obstacles at the same altitude as the ownship aircraft, (c) a shortest path function representation and the computed shortest path, (d) the objective function resulting from the Avoid Large Obstacle behavior. The example of Figure 6(a) shows a Digital Terrain Elevation Database (DTED) map containing a large obstacle represented by elevated terrain directly in the path between the aircraft and the target waypoint. The position of terrain features or other large obstacles is extracted from database information pertaining to the position of large obstacles. A binary map is generated from the database information to represent the horizontal position of large obstacles for a range of altitudes relative to the ownship aircraft. In Figure 6(b), the binary map is generated for the altitude of the ownship aircraft to illustrate the problem in two dimensions. Dark areas represent the horizontal position of large obstacles at that altitude. In this example, large obstacles are terrain features and airspace restrictions extracted from the DTED information and an airspace boundary database, respectively. A shortest path function giving the shortest path distance between points in space and the target waypoint is computed and shown in Figure 6(c), where lighter areas indicate shorter path distances between points and the target waypoint. In this example, the location of points is defined in terms of latitude and longitude for illustration purposes, but in the Avoid Large Obstacle behavior, the location of points is defined in terms of latitude, longitude, and altitude to accommodate for possible altitude changes. If the target is fixed, the shortest path function needs to be calculated only once. The number of points used to calculate the shortest path function is limited by a maximum distance between the points and the ownship aircraft, by a maximum shortest path distance between the points and 7 (d)

8 the target, and by a maximum number of points. The Avoid Large Obstacle behavior is improved by increasing travel time on points located on the edge of large obstacles in order to avoid issuing guidance commands that would cause the aircraft to travel too close to the edge of large obstacles. Once the shortest path function is calculated, the Avoid Large Obstacle behavior generates an objective function constructed to privilege heading, speed, and altitude values which set the aircraft closer to the target and away from large obstacles. It may be determined so as to assign higher arbitrary values to high-level commands that minimize the detour distance over the shortest path distance caused by maneuvering around large obstacles. The objective function generated by the Avoid Large Obstacle behavior is then combined with the objective functions generated by other behaviors. E. Avoid Collision Behavior Additional behaviors may be added to the guidance method such as an Avoid Collision behavior, which is intended to replace other behaviors in certain situations. The objective function generated by the Avoid Collision behavior is constructed to privilege high-level commands that maneuver the aircraft away of imminent collision threats and is determined so as to assign higher arbitrary values to high-level commands which maximize the distance at CPA between the aircraft and the obstacle(s) representing a threat. F. Follow Regulations Behavior Another behavior worth mentioning is the Follow Regulations behavior. The objective function generated by the Follow Regulations behavior is constructed to privilege high-level commands which maneuver the aircraft in a manner consistent with Federal Aviation Regulations (FAR) Part and other air traffic regulations effective in the airspace in which the aircraft is operating. It is determined so as to assign higher arbitrary values to high-level commands which, according to the situation, comply with air traffic regulations. The Follow Regulations behavior is itself comprised of several behaviors depicting applicable situations that depend on the encounter scenario (e.g., approaching head-on, converging, overtaking, descending, or ascending), the intruder aircraft category (e.g., balloon, glider, airship, rotorcraft, or airplane), and its priority (e.g., emergency, or minimum fuel), if this information is available. ADS-B is well suited for applying this behavior, as the information obtained from other ADS-B equipped vehicles includes an emitter category field and an emergency/priority code. IV. Results and Future Work Simulated flight tests were conducted in a software-in-the-loop (SWIL) test-bed in real-time. These tests represented a preliminary validation of the algorithms and allowed for the tuning and refinement of the behaviors. All behaviors were successfully demonstrated for a range of encounter scenarios involving one or multiple intruder aircraft. To better assess the reliability and efficiency of the SAA algorithms, a battery of encounter scenarios will be tested, which involves the implementation of Monte Carlo simulations in an environment that models the flight of aircraft in the NAS. Parallel to this assessment, the system will be progressively ported to a hardware-in-the-loop (HWIL) testing environment. Upon validation of these steps, flight testing on a small UAS will be initiated, and will involve flying the USA in near mid-air collision encounter scenarios with other aircraft. Acknowledgments This research was supported in part by Department of Defense contract number FA C-C006, Unmanned Aerial System Remote Sense and Avoid System and Advanced Payload Analysis and Investigation, and the North Dakota Department of Commerce, UND Center of Excellence for UAV and Simulation Applications. The authors would also like to acknowledge to contributions of the UASE laboratory team at UND. References 1 FAA Reauthorization Act of 2009, H.R.915, February 9, ADS-B Frequently Asked Questions, FAA Aviation Safety, URL: media/ads-b_faq.pdf [cited 3 March 2009]. 3 Borrelli, G. S., ADS-B to LINK-16 Gateway Demonstration: An Investigation of a Low-Cost ADS-B Option for Military Aircraft, The MITRE Corporation, Case # , Bedford, MA Benjamin, M. R., The Interval Programming Model for Multi-Objective Decision Making, Computer Science and Artificial Intelligence Laboratory, MIT, Cambridge, MA, Tech. Rep. AIM , September