The Hri-Cmu Corpus Of Situated In-Car Interactions

This paper introduces the HRI-CMU Corpus of Situated In-Car Interactions, a multimodal corpus of human-human interactions collected within highly sensored vehicles. The corpus consists of interactions between a driver and copilot performing tasks including navigation, scheduling and messaging. Data was captured synchronously across a wide range of sensors in the vehicle, including, near-field and far-field microphones, internal and external cameras, GPS, IMU, and OBD-II devices. The corpus is unique in that it not only contains transcribed speech, annotation of dialog acts and gestures, but also includes grounded object references and detailed discourse structure for the navigation task. We present the corpus and provide an early analysis of the data contained within. The initial analysis indicates that discourse behavior has strong variation across participants, and that general trends relate physical situation and multi-tasking to grounding behavior.