Use this URL to cite or link to this record in EThOS: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.716678
Title: Investigating interactional issues of automated planning support for disaster response
Author: Jiang, Wenchao
ISNI:       0000 0004 6352 0193
Awarding Body: University of Nottingham
Current Institution: University of Nottingham
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Abstract:
This thesis contributes to the understanding of the potential socio- technical issues that can emerge from the interaction between dis­aster responder teams and automated planning support, which in turn, leads to design implications for dealing with the emerged issues. Recently, natural and man-made disasters in Haiti, Chile and Japan drew the attention of researchers of disaster management systems. A lot of efforts have been made to study the technologies that can assist human responders to improve their performance. In the Disaster Re­sponse domain, a Disaster Response team, which typically contains several incident commanders and field responders, are faced with the problem of carrying out geographically distributed tasks under spa­tial and temporal constraints in a quickly changing task environment. Recent advances in multi-agent technologies lead to the possibility of building planning support for team coordination by automating the task planning process. However, it is unknown how the planning sup­port system can fit into the team organisation in a way that improves rather than hinders the team performance. This PhD work adopts a serious mixed reality game approach as the vehicle to explore socio-technical issues in complex real world settings. We developed AtomicOrchid, an emergency response game to create a task setting which mirrors aspects of real world Disaster Response operation. In the game trials, participants are recruited to play as field responders and incident commanders to carry out res­cue missions. Participants' experiences are observed and recorded as they coordinate with each other to achieve game objectives, with the support from an intelligent planner agent. The observations gathered in three AtomicOrchid field trials reveal how the organisational work is organised with and without automated planning support; how the guidance from the planner supports/interrupts natural human work- flow; how the human mediation (between planner and responders) played out to support the team work for better or worse. In the first study, field responders and Headquarters (HQ) coordi­nate without support of the intelligent planner. The result showed the team planning is dominated by local coordination between field players in a "situated" manner. The HQ is observed to successfully provide situational awareness to the field teams through remote mes­sages. However, the HQ has little direct influence on the planning and actions of field teams. In the second study, an automated planner was introduced to guide field responders directly without involvement of incident commanders. At its best, the agent is observed to take over routine planning activities while the humans focus on other is­sues such as finding teammates, targets and choosing the best routes. However, there also evidence showing the agent planning occasion­ally interrupts the workflow of the human team. In the third study, the system is modified to support incident commanders mediating task planning activities for field responders and the planning agent. We find that the human coordinator and automated planner agent can successfully work together in most cases, with human coordina­tors inspecting and 'correcting' the agent-proposed plans. However, occasional failures of planning are also observed. Through the three field studies, we find that highly automated planning can disrupt smooth human coordination potentially because it fails to consider the social cost of team and task allocation. There are evidences showing that human involvement in the planning loop can reduce the disruptions. However, in order to effectively involve human in the planning loop, a set of issues needs to be addressed, in­cluding: complacency effects; silent, missing or invisible information held by automated planner; and limited support for situated actions. The work also further generalises implications for interaction design, around the themes of balancing labour division, achieving common ground, facilitating accountability, and supporting situated actions.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.716678  DOI: Not available
Share: