Mixed-initiative Manned-unmanned Teamwork Using Coactive Design and Graph Neural Network

Mixed-initiative decision making is a flexible and effective way for coherent manned-unmanned teamwork (MUT). It allows the autonomous adjustment of levels of autonomy (LOA) and the human-robot collaboration modes according to the task requirements as well as the states of environments, robots and the human operator. However, it is still difficult for humans and robots to understand each other's intentions and motivations due to the challenges of cognition representation and behavior reasoning. In this paper, we propose a novel mixed-initiative MUT approach using coactive design and graph neural networks (GNN) towards explainable human-robot collaboration. First, an interdependence analysis table is designed for a specific manned-unmanned aerial vehicle task following the coactive design principles of observability, predictability and directablity. Then, a multi-agent dynamic task assignment system is designed based on a task model with key decision-making points. Finally, we have used the Graph Network Library to design a GNN model for adjusting the LOA among the MAV and UAVs in the given task.