An Agent-Based Approach to Aircraft Conflict Resolution with Constraints

Abstract

In recent years, much attention has been focused on emerging concepts of decentralized control for the National Airspace System. The introduction of a distributed decision-making environment in a future air traffic management system gives airspace users (pilots and airline operations centers) freedom to change and optimize flight plans in real time, in contrast to the current operational paradigm characterized by centralized decision-making by the air traffic service provider. Such an approach to air traffic operations will require individual decision-makers within the airspace to identify and solve, in real time, routing problems to avoid traffic conflicts, weather cells and Special Use Airspace. In this paper, we apply a modeling and simulation based approach to investigate the potential for a Principled Negotiation-based approach to the distributed air traffic management problem. Another objective is to identify the frequency with which truly collaborative (i.e., negotiation) behavior may be required to solve air traffic conflicts while avoiding regions of airspace. A free flight simulation in Los Angeles Center airspace, based on real traffic and Special Use Airspace data indicates that over 8% of aircraft-to-aircraft conflicts require negotiation. Principal Scientist; kharper@cra.com. Member, AIAA. Author to whom all correspondence should be addressed. ✝ Software Engineer. Member, AIAA. ‡ Software Engineer. § Senior Scientist. Member, AIAA Research Scientist, Automation Concepts Research Branch; Mail Stop 210-10; E-mail: kbilimoria@mail.arc.nasa.gov. Associate Fellow, AIAA. ✝✝ Software Engineer, Raytheon ITSS. Introduction Emerging concepts for future air traffic management (ATM) systems and procedures will dramatically change human roles and tasks in the National Airspace System (NAS). There are several technical initiatives in the ATM community for the development of new concepts of operations to meet the projected airspace demands of the future, while maintaining the overall safety of the NAS. The Free Flight paradigm (RTCA, 1995) gives aircraft operators the freedom to optimize their trajectories in real time, while requiring them to assume responsibility for maintaining safe separation from other aircraft and conforming to any ATM restrictions imposed by the air traffic service provider. One possible approach to implement Free Flight is the Distributed Air/Ground Traffic Management (DAG-TM) concept of operations (NASA, 1999). DAG-TM is characterized by distributed decision-making among pilots, air traffic service providers and airline operational control (AOC) personnel, who work cooperatively to optimize both individual and global operations, while maintaining safety as the highest priority. The introduction of this envisioned distributed decision-making (DDM) environment, however, will have profound implications on pilot/controller/AOC information requirements, roles and responsibilities, allocation of workload, communication, and decisionmaking throughout the ATM system. In order to evaluate the feasibility of such ATM concepts, there is a need to model the new rules and protocols governing individual behavior of the key decision-makers in the air transportation system, including pilots, air traffic controllers and airline dispatchers. Because these decision-makers will ultimately drive overall ATM system performance and safety, any modeling and simulation approach to system analysis must include realistic human behavior representations (HBRs) of the key decision-makers. Furthermore, if we are to capture AIAA Guidance, Navigation, and Control Conference and Exhibit 5-8 August 2002, Monterey, California AIAA 2002-4552 Copyright © 2002 by the author(s). Published by the American Institute of Aeronautics and Astronautics, Inc., with permission. 2 American Institute of Aeronautics and Astronautics realistic behaviors in the DDM environment, these HBRs must incorporate the underlying perceptual and cognitive capabilities and limitations that determine decisions made by these key players. To support these objectives, this work has focused on the development of an agent-based modeling approach of human behavior, which allows us to investigate the feasibility of distributed control through principled negotiation (Jacolin and Stengel, 1998; Wangermann and Stengel, 1996) in the air traffic environment, and provides a test bed for new communications protocols, information requirements, and traffic management models. The principled negotiation approach to distributed decision-making involves the search for solutions that provide the greatest mutual gains with respect to each participating individual’s requirements. Following from previous efforts (Harper et al., 1998; Harper et al., 1999), we have furthered the development of agent-based models of pilot, air traffic controller, and airline dispatcher behavior. The resulting agents are responsible for collaborating in a simulated advanced ATM environment to resolve traffic conflicts and avoid regions of airspace, e.g., convective weather cells or Special Use Airspace (SUA), in a safe, fair and timely manner. Under previous development efforts (Harper et al., 1999), we have demonstrated the capability of the agent-based models of collaborative decision-making to identify specific problems and solve them through democratic negotiation. In this paper, we are interested in investigating the actual requirements for truly collaborative decision-making in the ATM environment. Algorithmic conflict resolution and rerouting methods have been developed and demonstrated in the literature, and could be implemented within a real-time autonomous route planning and adjustment tool (Bilimoria, 2000; Krozel et al., 1996; Tomlin and Pappas, 1998). The question now becomes whether such fully decentralized methods are sufficient to support separation assurance and hazard avoidance, or if more complex distributed control strategies are required to maintain the efficiency and safety of the NAS. The paper is outlined as follows. First, we provide the technical background underlying the development of our agent-based models of human decision-making and collaboration through principled negotiation. We then present an overview of the knowledge engineering (KE) based model development process employed in the construction of agent-based representations of commercial pilots, air traffic controllers, and AOC dispatchers, followed by a summary of the results of the KE efforts. This discussion concludes in a description of the information processing, situation assessment, and collaborative decision-making functions developed across the three targeted agent models. We then present an overview of the integrated system architecture that brings these models together within the simulated air traffic environment provided by the Future ATM Concepts Evaluation Tool (FACET) (Bilimoria et al., 2001). To support the experimental objective of this paper, we then describe a simulated air traffic scenario that incorporates realistic traffic and SUA data for Los Angeles Center (ZLA). We populate this scenario with dynamic agent representations and analyze their collaborative problem-solving behavior to determine the level of complexity that can be expected in a distributed ATM environment, and the frequency and nature with which flight plan amendments must be truly negotiated among multiple decision-makers within the system. Finally, we present some conclusions. Human Behavior Representation Developing agent models of collaborative decisionmaking behavior among commercial pilots, air traffic controllers and AOC dispatchers are based on SAMPLE (Situation Assessment Model of Pilot-in-theLoop Evaluation), an agent-based human behavior modeling architecture originally developed to model fighter pilots, but designed within a domainindependent architecture. SAMPLE is based on a hierarchical definition of behavior, where the highest levels represent ongoing cognitive tasks and the lowest levels define specific information processing, situation assessment, and procedurally-driven decision-making models. We believe that SAMPLE provides a strong agent-based modeling approach for advanced ATM environments, supporting the modular development of various agent representations. SAMPLE Overview The SAMPLE human behavior model is based on the Rasmussen Hierarchy of human information processing and skilled behavior (Rasmussen, 1983), which provides a strong unifying theoretical framework for analysis of different human skills that may be modeled within a real-time simulation. By dividing skilled behavior into categories based on the degree of automaticity, complexity, and level of cognitive processing, this framework supports systematic skill decomposition and measurement of individual aspects of the overall skill on a part-task basis.