A Framework for Dynamic Task Allocation - Instantiated for Cognitive Task Load-based Adaptive Automation

Abstract

As technology advances, artificial agents such as robots are increasingly deployed to work on tasks in complex and dynamically changing environments. Often these sophisticated robots work together with human agents in a team. Because of these developments, the need for research into cooperation in mixed human-robot teams is increasing. An important aspect of cooperation is task allocation. Static task allocation is often not sufficient for dynamically changing environments, so dynamic task allocation is needed. In this thesis, a high-level framework for dynamic task allocation, aimed at improving team performance in mixed human-robot teams is presented. The framework details how context information can be used to find possible role assignments for actors and to evaluate these role assignments. The framework describes the important concepts in context information that influence team performance and can be used to dynamically allocate tasks. Secondly, the framework details how to use these role assignments with evaluation to find the optimal task allocation for a team. One of the important factors in context information is the cognitive task load of a human agent. Cognitive task load is an important predictor of human performance and is dependent on the tasks that are assigned to a human. The framework is used as a base for designing a model for adaptive automation. The model takes into account the cognitive task load of an operator and the coordination costs of switching to a new task allocation. Based on these two context factors it finds the optimal level of autonomy of a robot, separately for all tasks that need to be executed. This model is instantiated for a single human agent cooperating with a single robot in the urban search and rescue domain. A small experiment is conducted aimed at testing the model. Some encouraging results are found: the cognitive task load of participants mostly reacted to the model as intended. Furthermore, important focus points for improving the model are identified such as taking into account more context information, e.g. capabilities (human vs. robot) and preferences.