Human Spatio-Temporal Attention Modeling Using Head Pose Tracking for Implicit Object of Interest Discrimination in Robot Agents

摘要

Hazardous search missions are an excellent application domain for human-robot teams because cost-effective robot systems could be leveraged to reduce total mission duration time, improve search space thoroughness, and reduce human exposure to danger. Efficiently pairing robotic agents with human workers requires leveraging implicit communication when explicit techniques are either unavailable, socially unnatural, or impractical, such as is often the case with challenging search and rescue missions. Although successful implicit communication methods for robotics systems exist, (e.g., gestural recognition, activity recognition, and gaze management), a full suite of effective natural communication skills remains an open problem in robotics, thus preventing human-robot team solutions from being more commonplace. To help address this capability gap, we introduce a technique to implicitly model human spatio-temporal attention with a 3D heat map based on head pose trajectory tracking. We then show that this version of attention modeling can be applied by a robot agent to reliably extract Object of Interest (OOI) information for use in improving implicit communication in human-robot teams. This technique is evaluated in an OOI search task and a shared workspace clustered OOI discrimination task.