Teroos-touch: a context-aware shoulder-mounted telepresence robot mediating interactions via speech and social touch

This paper proposes Teroos-touch, a shoulder-mounted telepresence robot that adaptively selects between speech and social touch based on the local context. Remote operators often lack situational awareness, leading to two critical communication breakdowns: (i) spoken messages being missed in noisy or attention-demanding environments, and (ii) utterances becoming socially inappropriate or privacy-violating in the robot's location. Teroos-touch mitigates these challenges utilizing two Large Language Model (LLM)-driven reasoning modules. The Environmental Sense module decides whether to precede speech with a touch cue using ambient sound level and a scene description. The Social Sense module evaluates the appropriateness of the operator's utterance for the current situation; if deemed inappropriate, the system autonomously suppresses the utterance and replaces it with a touch cue. We evaluate this approach through three studies (laboratory, online video-based, and in-person wearable settings) and an exploratory case study. The results suggest that touch cues can improve noticeability and perceived context consideration in noisy settings, and that replacing socially inappropriate speech with touch can reduce embarrassment and improve likeability in selected settings.