NRI: FND: Spatial Patterns of Behavior in Human-Robot Interaction Under Environmental Spatial Constraints

Yale University, New Haven CT

Investigators

Abstract

This project promotes the progress of science and robotics by advancing autonomous reasoning about spatial patterns of group behavior during human-robot conversations. Spatial patterns emerge during conversations as a result of every participant's need to communicate while simultaneously perceiving everyone else's response. For example, circular conversational groups often emerge in open spaces. In physically constrained spaces, though, people's position might be influenced by nearby elements, such as walls and other people, leading to variations in acceptable group structure. To enable robots to cope with this variability, the project provides the empirical knowledge and methods needed to incorporate spatial constraints into the way robots reason about human (and robot) spatial formations. The project outcomes have implications across socially-relevant application domains in which user acceptance of co-robots can have a positive impact, including mobile service applications, education, and healthcare. Research activities will offer training opportunities to broaden participation in computing, serve to mentor and train future roboticists, and engage the public in the science of robotics. Building on foundational work in Human-Robot Interaction (HRI), this project addresses three main questions to advance perception and decision-making for co-robots in group settings: (1) how do spatial constraints influence conversational group formations in HRI; (2) how can robots detect these formations under spatial constraints; and (3) how can they autonomously generate appropriate spatial behavior to sustain conversations in constrained environments. To this end, this research will first focus on a formative study to better understand the effect of spatial constraints on group formations in HRI. This effort will result in a new public dataset of group-robot interactions that can be used to benchmark group detection approaches. The data will contribute to lowering barriers of entry to studying group human-robot interaction. Then, new methods for detecting spatial group formations in HRI will be developed by combining model-based and data-driven learning methods. Special consideration will be given to identifying groups in constrained environments. Finally, the project will investigate mechanisms to enable robots to take part in group formations under varying environmental spatial constraints. This last effort will help co-robots communicate with users and sustain group conversations by physically adapting to the environment. Together the outcomes of the project will help robots cope with the inherent complexity of multi-party interactions. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.