Coherence-Driven Multimodal Safety Dialogue with Active Learning for Embodied Agents

When assisting people in daily tasks, robots need to accurately interpret visual cues and respond effectively in diverse safety-critical situations, such as sharp objects on the floor. In this context, we present M-CoDAL, a multimodal-dialogue system specifically designed for embodied agents to better understand and communicate in safety-critical situations. The system leverages discourse coherence relations to enhance its contextual understanding and communication abilities. To train this system, we introduce a novel clustering-based active learning mechanism that utilizes an external Large Language Model (LLM) to identify informative instances. Our approach is evaluated using a newly created multimodal dataset comprising 1K safety violations extracted from 2K Reddit images. These violations are annotated using a Large Multimodal Model (LMM) and verified by human annotators. Results with this dataset demonstrate that our approach improves resolution of safety situations, user sentiment, as well as safety of the conversation. Next, we deploy our dialogue system on a Hello Robot Stretch robot and conduct a within-subject user study with real-world participants. In the study, participants role-play two safety scenarios with different levels of severity with the robot and receive interventions from our model and a baseline system powered by OpenAI's ChatGPT. The study results corroborate and extend the findings from automated evaluation, showing that our proposed system is more persuasive and competent in a real-world embodied agent setting.

Person Transfer in the Field: Examining Real World Sequential Human-Robot Interaction Between Two Robots

With more robots being deployed in the world, users will likely interact with multiple robots sequentially when receiving services. In this paper, we describe an exploratory field study in which unsuspecting participants experienced a ``person transfer'' -- a scenario in which they first interacted with one stationary robot before another mobile robot joined to complete the interaction. In our 7-hour study spanning 4 days, we recorded 18 instances of person transfers with 40+ individuals. We also interviewed 11 participants after the interaction to further understand their experience. We used the recorded video and interview data to extract interesting insights about in-the-field sequential human-robot interaction, such as mobile robot handovers, trust in person transfer, and the importance of the robots' positions. Our findings expose pitfalls and present important factors to consider when designing sequential human-robot interaction.