AUTOFLY-Aid: Flight Deck Automation Support with Dynamic 4D Trajectory Management for Responsive and Adaptive Airborne Collision

1. Introduction
The Single European Sky ATM Research (SESAR) and the US NextGen programs envision that the ever increasing demand of air transportation will require new methods for ensuring safe and efficient airborne flight planning. Revolutionary concepts such a Free Flight means the basic air traffic controller tasks associated with short term conflict resolution between aircrafts can potentially reach an intractable level under dense airspace operations. One natural solution to this problem is to take some of the work off the controller by delegating separation responsibility to appropriately equipped flight decks. On such flight decks, real-time conflict detection and resolution technology should be provided as to aid the pilot with immediate to short-term flight planning in dense and rapidly changing airspace environment. This means that new methods need to be developed for ensuring adaptive and responsive 4D trajectory management.

A part of the aiding is to provide the pilot with not only solutions but also enhanced situation awareness of the complex airspace picture. There are currently a number of flight deck displays that provide pertinent data associated with conflict detection and resolution to the pilot. These data sources include systems such as Traffic Collision Avoidance System (TCAS), Enhanced Ground Proximity Warning System (EGPWS), Aircraft Communications Addressing and Reporting System (ACARS), and Weather Radar and Enhanced Vision System (EVS). As of today, these multiple systems carry only a primitive hierarchical integration and force the aircrew to sort through each system while simultaneously interfacing with ground controllers [1]. This essentially means increased workload and a less than optimum situational awareness for the pilot as to identify and react towards potential hazards. One solution to this problem is to provide the pilot with one and consistent (and thus fully integrated and fused) air picture which embeds proximity with intent, This means that new methods need to be
developed to produce fused data with tractable data quality and integrity.

some_text

some_text

Airborne Collision Avoidance Systems (ACAS) and their implementations such as Traffic Collision Avoidance Systems (TCAS) are based on infrastructure and operations of ATM realm of the 20th Century. Specifically, in mid 1990s Traffic Alert and Collision Avoidance System (TCAS) was introduced to prevent mid-air collisions between aircraft. In TCAS II, in addition to TA, resolution advisories (RA) are introduced to instruct pilots on how to resolve conflict situations. In 2008, EUROCAE and RTCA have jointly revised operational standards of TCAS II, which is known as TCAS II version 7.1, to solve some safety issues that caused mid-air collisions. Even with current improvements, the primary shortcomings of TCAS can be summarized under 4 main themes. Specifically,

• TCAS (in operation) is limited to support vertical separation advisories.
• TCAS dynamic re-routing/re-advisory capability is limited to resolution advisory reversals, and in 
face of series of pilot blunders this limits the reliability of generated de-conflicts.
• TCAS does not incorporate weather, terrain, and ground and obstacle awareness and can potentially 
create advisories resulting in harming scenarios especially in close ground/terrain operation phases.
• TCAS does not provide resolution advisories in line with the aircraft’s “current” performance capability and flight envelope limitations.

In current ATM approaches the 4D trajectories means that trajectories are negotiated between the ground controllers and the aircraft FMS prior to departure, and as needed during flight, to ensure that the strategic flight plan remains conflict free. In many negotiation procedures, the ATM system uplinks time constraints to the aircraft, which computes in the FMS an accurate and flyable 4D trajectory and downlinks the trajectory to the ATM system for conflict probe. The negotiated trajectory is considered as a contract between the ATM provider and the aircraft and this should stay close this trajectory. In some of the proposed methods this strategy has the drawback of requiring too much throttle activity, reduced passenger comfort, increased fuel consumption, and increased engine wear. Then the strategy is relaxed by allowing the aircraft to be maintained within a specified tolerance of the negotiated trajectory and informing the controller if that cannot be achieved.

2. Literature Summary
With the new SESAR/NextGen air traffic capabilities and procedures, it is likely that the TCAS II threat detection and resolution logic will require modification to meet newly evolved operational requirements and traffic capacities. Due to the complexity of the logic, modifying the logic may require significant engineering effort [1]. The TCAS logic consists of several components: threat detection, initial sense selection, initial strength selection, and encounter monitoring and RA modification.

Mid-term and short-term decision makers (Air Traffic Control, pilots and TCAS) use different information sources, and they operate under different constraints and with different goals. In generally speaking, TCAS gets more accurate range or altitude infor- mation about an intruder than ATC, but TCAS cannot observe all the factors affecting traffic such as the location of hazardous weather, terrain, aircraft without transponders, or ATCo instructionsthis is the major reason that TCAS is certified to operate only as an advisory system [2].

MIT Lincoln Laboratory Air Traffic Monitoring Program group collected 200,000 flight hour (in 190 days) data within 60 nautical-mile coverage from from June 2005 to January 2006 [2]. The group observed 1725 RA events, resulted in that 9 RA events per day, or one RA in every 116 flight hours. By processing this data the following outcomes are proposed: Only 13% of pilot responses within 5 seconds and achieving a 1500 ft/min vertical rate (met the assumption used by TCAS). In 63% of the cases, the pilots maneuvered in the proper direction, but were not as aggressive as TCAS assumed. Pilots maneuvered in the opposite direction to the RA in 24% of the cases – some of these opposite responses are believed to be due to visual engaging with the intruder aircraft and the pilots decision that following the RA was not necessary. Opposite response to the TCAS RA can result in exactly the kind of mid-air collision happened at Uberlingen. Uberlingen accident, were posi- tively identified in which one aircraft flew opposite to its RA, a reversal did not occur[3].

A next generation air transport navigation systems should allow aircrafts to modify their flight plans during the flight without approval from a centralised control. There- fore Free Flight concept is extensively studied by research community including decen- tralised peer-to-peer conflict detection and avoidance systems. It is possible to integrate some free flight methods as to support the pilots with conflict resolution advisory (with pilot decision support systems). NextGen is currently investigating more delegation of traffic separation responsibility to the pilot [4, 5]. Early ASAS experiments showed promising results of assisted separation operations [6, 7] with the system where pi- lots are assisted in predicting and resolving loss of separation by cockpit automation, known generally as Airborne Separation Assistance Systems ASAS [8, 9].

Conflict detection algorithms should predict the picture of the future to issue an appropriate alert, and these methods are distinguished according to how they detect potential collisions. Three exploration methods have been identified in the literature and [2] also gives their definitions: nominal, worst case and probabilistic.

[10] proposes an open-loop solution involving 3-D single maneuver to be per- formed by the ownship without cooperation of intruder. The CDR algorithm in [11], namely NextCAS II, provides model-based solution that computes the alert thresholds not to violate of intruder’s protection zone. Proposed aircraft model includes certain protection zone and maneuver capabilities such as; climb/descent rate, turn rate, max-imum and stall speeds, and pilot response delay. The algorithm computes a set of six alert action with their alert times, then downs to single advisory. In [12], the Mixed In- teger Linear Programming (MILP), involving approximate model (point mass model) of dynamics with linear constraints, is applied to aircraft collision avoidance. [13] utilises the Mixed Integer Nonlinear Programming method for solving conflicts arising among several aircraft, but only velocity changing action is considered. [14] proposes an multi-layered open-loop ”almost blind engaging” method where the planner tries to solve ownship’s trajectory according to belief states of the intruder aircraft, and up- dates projected belief whenever new measurement information is arrived. The method exploits information uncertainty, communication delays and possible intruder action intents by using their probabilistic models.

Unlikely open-loop methods, instead of generating static plans at each information update, closed-loop methods [1, 15] generate an action sequence set that minimize cost by accounting future actions, and update likelihoods upon the new information avail- ability. Markovian Decision Process (MDP) solution, which is an open-loop solution, involves online and offline methods. In offline MDP, the algorithm needs to discreti- sation of the state space in order to efficiently approximate the value function within some error bound. Since computational complexity scales exponentially with the di- mension, where to apply to high-dimensional problems may be infeasible, [16] heads to solve the problem by only defining vertical motion. Online MDP solutions (such as in [17] involves Monte Carlo sampling) can be implement better to high dimensional state spaces but requires more computation. Online algorithms address the shortcom- ings of offline methods by only planning for the current belief state instead of planning for all possible situations. This methods are able to account for changes in the envi- ronment since they are executed once at each decision point. Online POMDP, namely Real-time POMDP, approach has been applied to aircraft collision avoidance problem in [18, 19]. A hybrid solution is also proposed in [19] where the expected utility of being in a particular belief state and selecting action is approximated as a function of the belief state computed online, and action utilities are computed offline. The method accounts for state uncertainty by integrating probabilistic models.

3. Research Objectives
Towards these goals, AUTOFLY-Aid will study “dynamic 4D trajectory management” to be implemented above the basic/passive TCAS solution using the on-board avionics and the SESAR enhanced flight deck situational awareness, coming from CNS (primarily ADS-B and its enhancements) and SWIM network. The “dynamic 4D trajectory management” is to be based on a hybrid and stochastic airspace model not only representing uncertainties associated with sensed and received airspace traffic and intent information, but also representing limitations associated with weather, terrain/obstacle and new conflict hazards. As an end result, the overall automation support system which embeds “dynamic 4D trajectory management” is envisioned to a) provide the pilots with alternative trajectories as tunnels-in-the-sky through avionics displays on the console and head-up displays in real-time, b) provide the flight crew with quantified and visual understanding of collision risks in terms of time and directions and countermeasures, and c) provide autonomous conflict resolution as an autopilot mode. Thus, ensuring highly responsive and adaptive airborne collision avoidance in face of ever challenging scenarios that involve blunders, weather/ terrain/ obstacle/ new conflict hazards. These algorithms and tools developed are currently being integrated on an Automation Support System for implementation on a Boeing 737 Flight Simulator with synthetic vision and reality augmentation.

some_text

4. Methodology
The main theoretical innovative and novel concepts as proposed and to be developed by AUTOFLY-Aid project are a) design and development of the mathematical models of the full composite airspace picture from the flight deck’s perspective, as seen/measured/informed by the aircraft flying in SESAR 2020, b) design and development of a dynamic trajectory planning algorithm that can generate at real-time (on the order of seconds) flyable (i.e. dynamically and performance-wise feasible) alternative trajectories across the evolving stochastic composite airspace picture (which includes new conflicts, blunder risks, terrain and weather limitations) and c) development and testing of the Collision Avoidance Automation Support System on a Boeing 737 NG FNPT II Flight Simulator with synthetic vision and reality augmentation while providing the flight crew with quantified and visual understanding of collision risks in terms of time and directions and countermeasures.

As an end result, the overall automation support system which embeds “dynamic 4D trajectory management” is envisioned to a) provide the pilots with alternative trajectories as tunnels in the sky through avionics displays on the console and head-up displays in real-time, b) provide the flight crew with quantified and visual understanding of collision risks in terms of time and directions and countermeasures, and c) provide autonomous conflict resolution as an autopilot mode. Thus, ensuring highly responsive and adaptive airborne collision avoidance in face of ever challenging scenarios that involve blunders, weather/terrain/obstacle/new conflict hazards.

Mathematical modeling of the airspace with uncertainties

“Real World” factors such as uncertainty in sensing, information, intent and rationality, asynchronous data and information flow with delays, equipment malfunctions, lack of centralized decision-making in short to immediate term collision avoidance, make responsive and adaptive airborne collision avoidance challenging. The problem is further complicated by the fact that the process is governed by humans and real aircraft dynamics (and thus with limitations of an aircraft and a human). In addition weather, terrain/ground and obstacle hazards, and new conflicts appearing in dynamically evolving scenarios lead to a potentially unbounded Airborne Collision Avoidance (ACA) problem complexity.

To design a conflict detection and resolution system, these factors should be considered in addition to increasing capacity demand and air congestion. Our dynamic modeling approach of the airspace hinges on hybrid systems methodology, which provides the framework for not only continuous dynamics but also discrete dynamics and logical jumps (and decisions). With the inclusion of stochastic processes and distributions, we model sensors, devices, information, intent, decisions and aircraft each with uncertainties and discrete/logical element under a coherent systems model. With regards to the representing aircraft dynamics, Mode Based Maneuver Automaton will be used. This finite state automaton cannot only represent the full dynamics and the limitations of the aircraft but also describe almost any maneuver with maneuver mode sequences. In addition, other aircraft’s intent is modelled through a stochastic risk-based decision model, which inherently captures all potential blunders and even irrational behaviour.

Besides geometric based localisation of terrain and obstacles, measurement/ information uncertainty and weather patterns are modelled through generic (and existing) stochastic sensor/information models and dynamic weather models respectively.

Real-Time probabilistic conflict detection algorithm

The conflict detection methodology is based on the idea of spatial search phenomena for potential conflicts including aircraft-to-aircraft conflicts and collisions with the obstacles such as severe weather patterns, terrain, and no-fly zones. This search method relies on creation of probabilistic flight trajectory (4DT) envelopes for the aircraft in local airspace for every predefined time window. These envelopes also include uncertainty factors existing in weather patterns and the flight models. The flight models naturally embed the stochastic nature in which the rationality and irrationality of the flight crews within the common airspace is presented with probabilistic action patterns. The main idea behind the Modal Maneuver Based PRM (Probabilistic Road Mapping) Planning is to divide an arbitrary flight maneuver into smaller maneuver segments (called maneuver modes – such as Level Flight Mode, Climb/Descent Mode, Lateral Loop Mode, Longitudinal Mode, Transition Mode and Roll Mode) and associate them with maneuver parameters (called modal inputs). The multi-modal maneuver search relies on a hybrid automaton, which chooses maneuvers from a finite maneuver set and then chooses their parameters from a continuous dynamically feasible region. This selection is made randomly in order to cover the whole flight envelope. The trajectory distribution map, which is the set of the generated maneuvers in a probabilistic distribution, represents all potential positions of the aircrafts in the future. If the generated 4D trajectory distribution maps conflict with ownship’s flight intent at high likelihood rates, this will serves as the alert for potential collision in a predefined unit time (or less).

some_text
some_text
some_text

Adaptive real-time probabilistic conflict resolution algorithm

4D conflict resolution methodology hinges on solving relaxed forms of the detected collision avoidance problem and then gradually refining the problem using the flight tracks of approximate solutions. This approach is implemented into two layers. In the first layer, Trajectory Planning Layer, the algorithm rapidly explores the airspace with an enhanced Rapidly Exploring Random Tree (RRT*), which includes asymptotic optimality property. This layer builds an approximate conflict-free will guide further steps of the algorithm. In the second layer, obstacle/collision free paths are connected with dynamic B-Spline curves. The approximation is further verified for collision and dynamic feasibility by computing the first and second derivatives of the spline in which correspond to the instantaneous velocity and acceleration. If the generated curve is not feasible, probabilistic repairing can be achieved by randomized waypoint (control point) placement on the B-spline curves iteratively and then the unit flight time is expanded to limit the acceleration within a controllable regime. Since B-Spline curves have a local support property, these repairing processes can be made on local path segments of interest without affecting the whole shape of the generated path. After obtaining the flight path with velocity history from the trajectory-planning layer, segment identification readily decomposes the flight path into a sequence of maneuver modes and its parameters. Mode-Based Maneuver Automaton implements this decomposition while ensuring transition rules for dynamic feasibility. This decomposition is used to define the proposed solution with basic maneuver identifications (such as climb, descent, lateral loop etc.) enabling generating standardized aural alerts (like TCAS) to give additional support to visual alerts. The generated conflict-free trajectory is also translated to visualization as a tunnel-in-the-sky to provide the pilot visual understanding of the generated solution.

Flight Deck Simulator integration and Decision Support tool technological demonstrations

These algorithms and tools to be developed are to be integrated on an Automation Support System for implementation on a Boeing 737 NG FNPT II Flight Simulator with synthetic vision and augmented reality. The augmented situational awareness is to be represented through console avionics displays (Synthetic Vision Display) and head‐up displays (Primary Augmented Flight Screen) in real‐time. The HUD/AR implementations to be tested include a) standard pilot-centered Head-up-Display (HUD), b) enhanced pilot-centered augmented reality goggles c) flight-deck centered Augmented Reality Screen Overlay Projection. Augmented decision aiding implementations will include new touch screen and haptic input devices/switches for moving through HUD and Augmented Reality Display pages and switching/choosing between alternative trajectories represented as tunnels- in-the-sky illustrations.

some_text

In our concept, this interaction with the aid/decision support system occurs in real-time while “flying”. This is in direct comparison to executing pre-negotiated and pre-planned trajectories. Specifically, the pilot can in real-time ignore a generated solution, and can either choose to skip current trajectory advisory (with page-down like haptic devices on yoke, or touch-screen gestures) for checking alternative solutions, or modify the solution by ignoring to follow the trajectory (augmented tunnel-in-the-sky). Once the flight plan is modified through actual implementation, the support system re-plans trajectory solution on-the-fly according to the current states. Following the proposed and visually demonstrated solution implies acceptance of the proposed solution and the pilot can also choose to send proposed trajectory to the autopilot for autonomous execution (through FMS link). Any deflection response (e.g. pulling stick) during the autonomous execution switches the system to self-execution state supported with advisories.

some_text

The proposed decision-support approach also addresses another issue in the human machine interfaces. In the persistent interfaces, the pilot gives her full attention to the interaction to keep track of the labels and pages visited in the process of discovering the correct action sequence. This tunneling effect results in degraded flight deck efficiency and reduced safety margins. In the analysis of the American Airlines Flight 965 accident at Cali, Columbia, it is reported that the flight crew spent an inordinate amount of precious time heads-down trying to manipulate the FMS to perform an infrequent task, which is not well supported by the interface, and this inefficient interaction contributed to the occurrence of the accident with several other circumstances. In our approach, this tunneling effect and persistent interaction has been diluted with lack of additional “channels to interact with”.

In planned functional architecture of the Flight Deck Simulator, the Traffic and Weather Generator (T&WG) is used as a scenario generator. Flight Simulator Software visualizes these generated scenarios. The Traffic and Weather Generator (T&WG) replays both artificially generated scenarios including air congestion and severe weather conditions, or collected live WX and ADS-B data (ITU CAL currently operates a ADS-B radar and Eurocontrol EGNOS DCN at Istanbul Ataturk Airport) transmitted from planes operating in Eastern European Airspace. The T&WG feeds the SWIM cloud in real-time, and any external Air Traffic Monitoring System or simulator can be connected to the flight-deck network by establishing connection to the SWIM. The Flight Deck Information Management (FDIM) system is a local in-cockpit data management system, which gathers all information (such as traffic, capacity, weather, terrain etc.) via communication avionics emulators; parses and broadcasts them to the their clients. For detected potential conflicts (including mid-air and ground), the Air Conflict Detection and Resolution (CDR) system generates dynamic conflict resolution to provide the pilots support with its alternatives and these 4D trajectories and corresponding obstacles are visualized through head-up-display (HUD), with Augmented Reality (AR) and Synthetic Vision display. Through the flight-deck pilot interface, the pilot can switch between alternative dynamically generated solutions via touch-screen, switches and haptic devices. In order to extend situational awareness of the pilot over entire flight operation, the Synthetic Vision display also offers additional pages including long-term, mid-term and short-term threat screens as explained.

References
[1] M. J. Kochenderfer and J. P. Chryssanthacopoulos, “Robust airborne collision avoidance through dynamic programming,” Massachusetts Institute of Technol- ogy, Lincoln Laboratory, Project Report ATC-371, 2011.
[2] J.E.KucharandA.C.Drumm,“Thetraffic alert and collision avoidance system,” Lincoln Laboratory Journal, vol. 16–2, p. 277, 2007.
[3] GFB-AAI, “Investigation report,” in BFU, German Federal Bureau of Aircraft Accident Investigation AX001-1-2/02, Braunschweig, Germany, 2004.
[4] D. Wing, “A potentially useful role for airborne separation in 4d-trajectory atm operations,” in 5th AIAA Aviation Technology Integration and Operations (ATIO) Conference, 2005.
[5] S. M. Green and K. D. Bilimoria, “Distributed air/ground traffic management for en route flight operations,” Air Traffic Control Quarterly, vol. 9, pp. 259–285, 2001.
[6] R. Barhydt and P. Kopardekar, “Joint nasa ames/langley experimental evalua- tion of integrated air/ground operations for en route free maneuvering,” in 6th USA/Europe Air Traffic Management R&D Seminar, 2005.
[7] MFFP, “Mediterranean free flight programme final report 2005,” in MFFP Tech- nical Report D821, 2005.
[8] K. Krishnamurthy, B. Barmore, and F. Bussink, “Airborne precision spacing in merging terminal arrival routes,” in 6th USA/Europe Air Traffic Management R&D Seminar, 2005.
[9] H. A. P. Blom, B. K. Obbink, and G. I. Bakker, “Safety risk simulation of an airborne self separation concept of operation,” in 7th AIAA Aviation Technology Integration and Operations (ATIO) Conference, 2007.
[10] G. Dowek, A. Geser, and C. Mun ̃oz, “Tactical conflict detection and resolution in 3-d airspace,” in NASA CR-2001-210853, 2001.
[11] R. Chamlou, “Design principles and algorithm development for two types of nextgen airborne conflict detection and collision avoidance,” in Integrated Com- munications Navigation and Surveillance Conference (ICNS), 2010, may 2010, pp. N7–1 –N7–12.
[12] A. Richards and J. How, “Aircraft trajectory planning with collision avoidance using mixed integer linear programming,” in American Control Conference, 2002. Proceedings of the 2002, vol. 3, 2002, pp. 1936 – 1941 vol.3.
[13] M. Christodoulou and S. Kodaxakis, “Automatic commercial aircraft-collision avoidance in free flight: the three-dimensional problem,” Intelligent Transporta- tion Systems, IEEE Transactions on, vol. 7, no. 2, pp. 242 – 249, june 2006.
[14] E. Koyuncu and G. Inalhan, “Exploiting delayed and imperfect information for generating approximate uav target interception strategy,” Journal of Intelligent & Robotic Systems, vol. 69, pp. 313–329, 2013. [Online]. Available: http://dx.doi.org/10.1007/s10846-012-9693-6
[15] M. J. Kochenderfer and J. P. Chryssanthacopoulos, “Partially-controlled markov decision processes for collision avoidance systems,” in International Conference on Agents and Artificial Intelligence, Rome, Italy, 2011.
[16] S. Temizer, M. J. Kochenderfer, L. P. Kaelbling, T. Lozano-pe ́rez, and J. K. Kuchar, “Collision avoidance for unmanned aircraft using markov decision pro- cesses,” AIAA Guidance, Navigation, and Control Conference, 2010.
[17] T. Wolf and M. Kochenderfer, “Aircraft collision avoidance using monte carlo real-time belief space search,” Journal of Intelligent & Robotic Systems, vol. 64, pp. 277–298, 2011. [Online]. Available: http://dx.doi.org/10.1007/s10846-010- 9532-6
[18] L. F. Winder and J. K. Kuchar, “Hazard avoidance alerting with markov deci- sion processes,” Massachusetts Institute of Technology, Lincoln Laboratory, Se- ries/Report no.: ICAT-2004-4, 2004.
[19] J. P. Chryssanthacopoulos, M. J. Kochenderfer, and J. Mykel, “Accounting for State Uncertainty in Collision Avoidance,” Journal of Guidance Control and Dy- namics, vol. 34, no. 4, pp. 951–960, 2011.