Human footsteps play a significant role in everyday life, allowing individuals to discern the emotions, gender, and intentions of others solely from the sound of their footsteps. However, the influence of footstep sounds made when walking on different floor types in virtual reality (VR) environments when joining conversational groups remains unclear. In this paper, we present a controlled study (N=50) to assess the impact of five different floor types, associated with specific footstep sounds and visuals, on the persuasiveness of Embodied Conversational Agents (ECAs) when inviting participants to join a free-standing conversational group. We analyze routes taken by participants and the positions at which they join the group, which may be compliant or not with the agent’s request when approaching the group while walking on different virtual floor types. Our findings reveal that the type of floor being walked upon, defined by footstep sounds and visual appearance, significantly impacts the persuasiveness of ECAs and the trajectories taken by participants to join the group. Participants took longer paths and joined the group in the presence of more pleasant footstep sounds. Further, they tended to adhere to social norms by avoiding walking through the group’s center.

This paper presents an evaluation of the use of large language models (LLMs) for grading clinical reasoning during rheumatology medical history virtual patient (VP) simulations. The study explores the feasibility of using state-of-the-art LLMs, including both general-purpose models, with various prompting strategies such as zero-shot, analysis-first, and chain-of-thought prompting, as well as reasoning models. The performance of these models in grading transcribed dialogues from VP simulations conducted on a Furhat robot was evaluated against human expert annotations. Human experts initially achieved a 65% inter-rater agreement, which resulted in a pooled Cohen’s Kappa of 0.71 and 82.3% correctness. The best LLM, o3-mini, achieved a pooled Kappa of 0.68 and 81.5% correctness, with response times under 30 seconds, compared to approximately 6 minutes for human grading. These results indicate the possibility that automatic assessments can approach human reliability under controlled simulation conditions while delivering time and cost efficiencies.