Poster: Phase-Based Latency Mitigation in LLM-Driven VR Agents
Response latency in Large Language Model (LLM)–driven Embodied Conversational Agents (ECAs) can disrupt conversational flow. We investigated whether phase-specific multimodal feedback can mitigate perceived waiting time without modifying the underlying dialogue pipeline. In a within-subjects study (n = 35), participants experienced feedback during user speech (USP), the response-waiting phase (RWP), both phases, or neither phase. Results showed that RWP feedback consistently reduced perceived latency, whereas USP-only feedback showed weaker and partly distracting effects, while perceived social qualities remained unaffected in all conditions. This highlights the importance of temporally aligned feedback for improving perceived responsiveness.
@inproceedings{10.1145/3821409.3833886,
author = {K{\"u}hlem, Konstantin W. and Zhou, Ying and Kuhlen, Torsten W. and B{\"o}nsch, Andrea},
title = {Phase-Based Latency Mitigation in LLM-Driven VR Agents},
year = {2026},
isbn = {9798400727993},
publisher = {Association for Computing Machinery},
address = {New York, NY, USA},
url = {https://doi.org/10.1145/3821409.3833886},
doi = {10.1145/3821409.3833886},
booktitle = {Proceedings of the 2026 ACM Symposium on Applied Perception},
articleno = {53},
numpages = {2},
keywords = {large language model, embodied conversational agents, response latency, turn-taking, multimodal feedback, virtual reality, user study},
series = {SAP '26}
}
