Coordination Mechanisms for Human-Robot Teams in Space

The goal of this research was to develop a computational framework for human-robot coordination in space environments. The framework was based on the concept of a Shared Mental Model (SMM), which is a distributed construct used in effective human teams to track various aspects of the team (e.g., role, spatial location, perspective, etc.) and the task (equipment, procedures, scenarios, etc.) to manage the interaction. The SMM framework includes knowledge representations of the above components (and others), as well as rules that use the representations to update the SMM state and to adapt robot behavior. By representing the same kind of task and team information used by humans, this framework allows for the formation and maintenance of common ground in the team. Moreover, the shared representations support interaction in a variety of team configurations, including teams that operate at varying spatial ranges, time scales, and interaction modalities. To properly evaluate the SMM framework’s ability to support team coordination requires: 1) implementation in a robotic architecture, and 2) a novel task environment in which a human-robot team performs an interdependent task in a space environment. To this end, we implemented the framework in the DIARC cognitive robotic architecture and developed an evaluation platform using a 3D simulated spacecraft environment in which humans and virtual robots interact in virtual reality (see Figure A). The virtual environment is scalable to any number of human and robot teammates, and can be customized to explore different properties of the team, e.g., structure, number/type of agents, interaction distance, etc. For an initial evaluation, we used a team structure involving one human and two onboard robots who communicate in natural language to perform a joint-maintenance task aboard a spacecraft. Using this configuration, we ran a human-subjects experiment to test the benefit of the SMM framework on team performance. Half of the participants were partnered with robots running the SMM framework, and the other half were in the non-SMM baseline condition, which involved robots with similar capabilities, but without the ability to access the shared knowledge representations in the SMM. We found that teams in the SMM condition performed significantly better in the task, and that this was due to the ability of the SMM framework to maintain accurate and updated knowledge that was shared between the robots (see Figure B). This represents the first empirical support for the hypothesis that SMMs improve performance in human-robot teams, and suggests that our framework is a promising approach to human-robot coordination. This work has the potential to be applied in the near-future to many of NASA’s objectives that involve heterogeneous human-robot teams. Some examples include: joint maintenance and repair on the ISS or LOP-G, site preparation, scientific experimentation, future missions to Mars, and other tasks for which such teams are needed. For each of these applications, the virtual environment can be modified accordingly, allowing for rigorous experimental evaluation of various team performance measures. Overall, this project demonstrates a novel and effective approach to improving coordination and performance in human-robot teams, bringing us a step closer to NASA’s vision of the joint-exploration of space.