Getting virtual agents into the Loop - An Adaption to human performance

Kurzfassung

Congestions at the airport increase the importance of an effective cooperation among the different airport management stakeholders. The airport management is specially challenged in situations where so called disruptive events (e.g. weather events, security alarms or union strikes) with the potential to affect the entire European air traffic system hit the airport. To minimize delay propagation to other airports, an efficient planning and use of the existing resources is required. Besides planning the individual resources, each stakeholder needs to adjust his plans with the other participants. Fulfilling the individual planning tasks can already be very time consuming. Conducting additional cooperation might lead to an overload of the stakeholder. Virtual agents have the potential to take over some of the stakeholders tasks and thereby relieve the stakeholder. For this purpose the German Aerospace Center (DLR) developed the virtual agent CLAUDI (Collaborative Learning Assistant for User Defined Interactions). CLAUDI is capable of conducting airport and airline planning tasks. Moreover, CLAUDI cooperates via chat system with multiple airport stakeholders and conducts the planning tasks. Various validations performed as Human-In-the-Loop simulations evaluated and proved the potential of CLAUDI. Nevertheless the cooperation with CLAUDI was examined to be challenging for the stakeholders. Writing and answering with the full speed of computational power lead to misunderstandings. Replies to questions were partially unrecognized as they were not immediately expected by the human partners. Moreover, very long messages being sent instantaneously were not fully read. This paper presents the adaption of the virtual agent CLAUDI to the limitations and characteristics of human processing and communication. For this purpose the reading, decision making and writing of CLAUDI was adapted to the speed of a human process. This adaption was evaluated in a validation with 18 participants. The validation uses the airport management simulation of the German Aerospace Center to challenge a team of four participants with several disruptive events. The role of the stand and gate allocation manager is taken over by either a human actor or a virtual agent. The results of the simulation runs with the human actor are compared with those of the virtual agent. Based on the analysis of the results it will be shown that the reduction in communication and decision speed leads to a better cooperation with CLAUDI.