Evaluating Zero-Shot and One-Shot Adaptation of Small Language Models in Leader-Follower Interaction
arXiv:2602.23312v3 Announce Type: replace-cross Abstract: Leader-follower interaction is an important paradigm in human-robot interaction (HRI). Yet, assigning roles in real time remains challenging for resource-constrained mobile and assistive robots. While large language models (LLMs) have shown promise for natural communication, their size and latency limit on-device deployment. Small language models (SLMs) offer a potential alternative, but their effectiveness for role classification in HRI has not been systematically evaluated. In this paper, we present a benchmark of SLMs for leader-foll
The proliferation of AI and robotics necessitates more efficient and deployable language models for real-time interaction, especially as hardware capabilities improve for smaller form factors.
Evaluating Small Language Models (SLMs) for human-robot interaction directly addresses the challenge of deploying advanced AI capabilities on resource-constrained devices, broadening the applicability of AI agents in physical environments.
The demonstrated effectiveness of SLMs could enable a new wave of autonomous robotic systems with sophisticated interaction capabilities without relying on large, cloud-based language models.
- · SLM developers
- · Robotics companies
- · Edge computing providers
- · Assistive technology manufacturers
- · Companies reliant solely on large language models for HRI
- · Cloud infrastructure providers for real-time HRI
Increased development and deployment of SLM-powered robots for diverse applications.
Reduced latency and increased privacy for human-robot interactions due to on-device processing.
The development of a competitive ecosystem for SLM hardware and software optimized for specific robotic tasks, potentially creating new industry standards.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG