M-CoDAL is a multimodal-dialogue system that enables robots to better understand and communicate in safety-critical situations by leveraging discourse coherence relations for improved contextual understanding.

• Uses active learning to continuously improve safety responses
• Interprets visual cues alongside verbal communication
• Enhances robot ability to handle safety violations in real-world environments
• Creates more coherent and contextually appropriate responses in hazardous situations

This research significantly advances security capabilities for embodied AI by improving how robots identify, communicate about, and respond to potential safety hazards during human-robot interaction.

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