Wi-Fi sensing adapts better across users and environments with AI agent
CSI-Agent: LLM-Assisted Few-Shot Adaptation for Cross-Domain Wi-Fi CSI Sensing
Artificial IntelligenceNetworking and Internet Architecture
Summary
Wi-Fi signals can be used to recognize human activities without devices, but current systems often fail when used with new people or in different places. The authors propose CSI-Agent, an AI-powered helper that uses limited new examples to decide the best way to adapt sensing models for each activity class. Instead of looking at raw signals directly, CSI-Agent gathers evidence about how well each activity is predicted and chooses whether to keep or change the default method, improving accuracy. Tests on multiple datasets show that this approach boosts performance significantly compared to existing methods.
What this means in practice
- •For smart home developers: Enhance device-free activity recognition performance across varying household users and environments using few-shot adaptation powered by an AI planning agent.
- •For security system engineers: Improve reliability of Wi-Fi-based motion and presence detection when deployed in diverse locations without extensive labeled data.
Authors
Tianya Zhao, Chuan Liu, Xuyu Wang
Abstract
Wi-Fi channel state information (CSI) has enabled device-free sensing applications such as human activity recognition. However, CSI sensing models remain brittle in cross-domain deployment, where changes in users or environments can produce incorrect predictions. Existing solutions usually treat this problem as an offline model-design problem, by pretraining a stronger representation or applying one fixed adaptation method to the entire target domain. In practice, labeled target data are scarce and different classes may fail in different ways under the same domain shift. To address this, we propose CSI-Agent, an evidence-seeking LLM agent that reformulates cross-domain CSI adaptation as a deployment-time decision-making problem. Rather than processing raw CSI or making sample-level predictions, CSI-Agent summarizes target-domain behavior into sensing-grounded class-level evidence. It establishes a strong target-adaptive default from complementary CSI views and uses an LLM planner to determine whether each class should retain the default or invoke a specialized action. Deterministic verification and bounded execution further reduce unreliable interventions. We evaluate CSI-Agent on four public datasets using five cross-domain splits covering device, user, environment, and compositional shifts. Under 1-shot adaptation, CSI-Agent achieves the best target-domain performance across all splits and improves the average Macro-F1 by about 16\% compared to the strongest baseline method.