 Andrea Bonzano and Pádraig Cunningham ECAI 98. 13 th European Conference on Artificial Intelligence Edited by Henri Prade. Published 1998 by John Wiley and Sons Ltd. Hierarchical CBR for Multiple Aircraft Conflict Resolution in Air Traffic Control 1 Andrea Bonzano and Pádraig Cunningham 2 1 This research has been funded by Eurocontrol Experimental Centre, Paris 2 Dept. of Computer Science, Trinity College, Dublin, Ireland. Abstract. In this paper we present a Case-Based Reasoning system that helps air traffic controllers to solve aircraft conflicts. In particular, we focus on the hierarchical aspect of the CBR system which is able to solve multiple aircraft conflicts, i.e. conflicts that involve three or more aircraft. It is not practical to build a case-base for the different multiple aircraft possibilities as has been done for two aircraft conflicts. Instead we explore the possibility of using case fragments from two aircraft conflicts in multiple aircraft situations. The hierarchical structure that we describe here makes this possible. This involves the use of some high-level analysis of the solutions coming from the case base because the solution to a multiple aircraft conflict is not necessarily one of the solutions of the component two aircraft conflicts. The hierarchical structure allows the use of the same case-base for both two aircraft conflicts and multiple aircraft conflicts with big savings in space and time. 1 INTRODUCTION Case-Based Reasoning (CBR) has emerged from research in cognitive psychology as a model of human memory and remembering. It has been embraced by researchers of AI applications as a methodology that avoids some of the knowledge acquisition and reasoning problems that occur with other methods for developing knowledge-based systems. It is intuitively appealing because people readily accept that much of human competence is based on the reuse of solutions to previously encountered problems. However the reality of CBR falls short of the flexibility of reuse that humans can achieve. The standard CBR structure provides a restricted form of reuse where the new problem needs to be isomorphic to a problem in the case-base. In this paper we explore how a hierarchical approach to CBR [1-5] can allow for more flexible case reuse. We describe a hierarchical extension to ISAC (Intelligent System for Aircraft Conflict Resolution), a CBR system that helps air traffic controllers to solve conflicts between sets of aircraft [6]. ISAC assists the controllers in the first two stages of the conflict resolution process: the selection of the aircraft to manoeuvre and decision of the type of manoeuvre. The third stage, the specification of the details of the manoeuvre, is left to the controller. While it has been possible to build a case that provides adequate cover for two-aircraft conflicts (TACs) it is not practical to provide dedicated cases to cover the different types of multiple aircraft conflicts that can arise. For instance, the third aircraft in a three-aircraft conflict increases the number of possibilities by an order of magnitude. The hierarchical structure that we describe here allows for the reuse of TAC case components in solving MACs. This greatly extends the reusability of cases and makes the solving of MACs practicable using a case-based approach. Before describing the hierarchical extensions to ISAC we present an overview of air traffic control and multiple aircraft conflicts in section 2. In section 3 we describe the case representation used in ISAC and in section 4 we present the possibilities for hierarchical CBR in air traffic control. In section 5 we work through and example of how ISAC can solve a three aircraft conflict and present the results of a preliminary evaluation of the system before concluding in section 6. 2 MULTIPLE AIRCRAFT CONFLICTS Internationally agreed rules exist defining separation standards below which two aircraft are said to be "in conflict". Minimum horizontal separations are typically 5 nautical miles (1nm = 1852m) in radar controlled regions and either 1000 ft or 2000 ft vertically, depending on altitude. In areas not covered by the radar the horizontal separation is bigger. In practice, controllers will often apply separations significantly larger than 5 nautical miles or 1000 ft, mainly due to the difficulties they have in accurately visualising future trajectories and conflict situations. In [7], three types of conflict sets have been identified as being the most common: • one versus one: the two conflicting aircraft are isolated from other conflicts; • one versus two: two separated conflicts sharing a common aircraft; • three-at-once: three conflicts among three aircraft. Two Aircraft Conflicts (TACs) are already efficiently solved by the non-hierarchical version of ISAC [6]. The performance of this version could be increased with an improvement of the case-base. In a Multiple Aircraft Conflict (MAC) the situation is more complex. If a MAC is decomposed into TACs, there is the risk of solving the wrong pair first. An overall view is necessary to decide which aircraft has to be manoeuvred. B A C C B A Figure 1: Types of Multiple Aircraft Conflicts.