Implicit Communication of Contextual Information in Human-Robot Collaboration
Yan Zhang
TL;DR
This work investigates implicit communication in human-robot collaboration by framing implicatures as contextual cues that can be inferred and generated by robots. It adopts a three-phase research plan: (1) study how human implicatures affect collaboration, (2) examine robots' implicit backchannels and proactive cues and how they should adapt to humans, and (3) develop a cooperative multimodal multi-LLM system that learns from human implicit communication. Empirical results from a lab study with TIAGo show that robots capable of interpreting implicatures can enhance perceived team performance, trust, and anthropomorphism, though suitability is task- and context-dependent. The proposed multi-LLM architecture aims to mitigate issues like hallucinations and to enable bidirectional, natural HRC, with potential impact on healthcare, manufacturing, and home environments.
Abstract
Implicit communication is crucial in human-robot collaboration (HRC), where contextual information, such as intentions, is conveyed as implicatures, forming a natural part of human interaction. However, enabling robots to appropriately use implicit communication in cooperative tasks remains challenging. My research addresses this through three phases: first, exploring the impact of linguistic implicatures on collaborative tasks; second, examining how robots' implicit cues for backchanneling and proactive communication affect team performance and perception, and how they should adapt to human teammates; and finally, designing and evaluating a multi-LLM robotics system that learns from human implicit communication. This research aims to enhance the natural communication abilities of robots and facilitate their integration into daily collaborative activities.