Research Abstracts - 2007

Introduction

In settings such as space, military, aviation, and medical industries, teams of people must effectively coordinate to accomplish complex tasks involving ordering, timing, and resource constraints. These tasks are often performed under stress induced by uncertainty, ambiguity, and time pressure. As robots are increasingly introduced into these domains, it is important to explore how a robot can most naturally coordinate with a human to perform these types of complex tasks under stress.

Literature on effective human coordination strategies [1-3] provide an excellent basis for understanding how a robot can most naturally coordinate with a human to perform complex tasks under stress. The robot should be able to:

- act in anticipation of the human’s actions by providing information or other needs of the human without explicitly being requested to do so.

- incorporate into its action planning the information and other resources that the human provides to the robot in anticipation of the robot’s actions.

- dynamically adjust to redistributions in workload to make use of idle periods.

- Avoid relying entirely on the human to explicitly communicate task coordination information (for example, “robot: do this next”). Instead the robot should be able to interpret status information about the task, environment, or human’s actions (for example, “robot: I have finished subtask A”) and incorporate these into its own action planning.

The goal of my research is to develop a tractable method allowing a robot to use these strategies to naturally coordinate with a human, and then demonstrate effective human-robot coordination on a hardware platform.

Previous Work

The reader is referred to [4] for review of progress towards simulating human-teamwork behaviors with agent-only and mixed human-agent teams. While none of these applications are specifically tailored to human-robot coordination of complex tasks under stress, they each tackle significant elements of the problem. However, three areas remain largely unexplored: (1) extended action planning for complex tasks involving ordering and timing constraints, (2) dynamic redistribution of workload for tasks involving ordering and timing constraints, and (3) the role of implicit communication behaviors in informing effectual (1) and (2).

Innovative Approach

We assume that, from the perspective of the robot, the human’s choice of actions and the precise durations of actions are uncontrollable. We also assume that the robot is able to control its own choice of actions and the precise duration of actions. We then model the collaborative task as a temporal plan network with uncertainty (TPNU) to capture ordering and timing constraints. Inputs include a sequence of the human’s actions over time, and information (in the form of implicit cues) the human offers to the robot in anticipation of the robot’s actions. The objective is to produce a dynamic control policy for the robot to support the human in completing the task. The control policy consists of the robot’s action choices and duration of actions, and information (in the form of implicit cues) the robot offers to the human in anticipation of the human’s actions.

References:

[1] Serfaty, D., Entin, E., and Deckert, J. (1993). Team adaptation to stress in decision making and coordination with implications for CIC team training (Report No. TR-564, Vol. 1 & 2) Burlington, MA: ALPHATECH.

[2] Kleinman, D., and Serfaty, D. (1989). Team performance assessment in distributed decision-making. In Proceedings of the Interactive Networked Simulation for Training Conference, pp. 22-27. Orlando, FL.

[3] Orasanu, J. (1990). Shared mental models and crew decision making (CSL Report No. 46), Princeton, NJ: Princeton University, Cognitive Science Laboratory.

[4] Fan, X., and Yen, J. (2004). Modeling and simulating human teamwork behaviors using intelligent agents. Physics of Life Review, Vol. 1, pp. 173-201.