Indoor radio map estimation (RME) is critical for wireless tasks such as coverage planning, localization, and network deployment. Existing approaches mainly rely on supervised learning, which demands costly labeled data and task‑specific model training, while indoor settings contain privacy‑sensitive information like building layouts that cannot be centralized. This work formulates indoor RME as a collaborative in‑context learning (ICL) problem and introduces the Poly‑Ripple framework. Multiple clients perform ICL with large autoregressive vision models (LAVMs) without updating the model; each client is prompted by its private demonstrations and predicts the radio map for the same scene query. To capture the causal propagation from the transmitter (Tx) outward, ripple‑ordered autoregressive modeling is adopted. An importance‑aware aggregation scheme is further proposed to exploit heterogeneous contextual knowledge distributed across clients under a limited context length. The paper provides theoretical analysis and extensive experiments on a public indoor RME benchmark. Results show that Poly‑Ripple consistently outperforms representative ICL baselines and aligns with the theoretical predictions.
Review