CPPU: Policy Space Diversity for Informative Path Planning and GAI-Enabled Updating CSI in ISAC

Abstract

Integrated Sensing and Communication (ISAC) is a core component of 6G technology, driving advancements in networking and communication in combination with Generative Artificial Intelligence (GAI). By utilizing GAI’s predictive functions, indoor positioning achieves higher efficiency through optimized fingerprint localization methods. Traditional approaches to Channel State Information (CSI) collection remain costly and lack adaptability in dynamic environments. To address these challenges, this paper proposes an algorithm that combines multi-agent learning with a GAI model for CSI acquisition, Path Planning, and Updating, namely CPPU. Initially, the terrain is partitioned into multiple regions, each corresponding to a specific agent, ensuring comprehensive path coverage and eliminating backtracking solutions. Subsequently, the dynamic programming strategy of the policy space response oracle is employed to identify informative paths within each region. These paths are determined by integrating full-coverage paths with a small dataset of real CSI, thus representing the most informative paths that reflect the distribution of CSI. CSI data along these informative paths is then collected. Finally, the GAI model is deployed to predict and update the CSI distribution at the remaining points. Experimental validation in two real-world scenarios verifies that CPPU significantly reduces data acquisition costs while maintaining localization accuracy comparable to existing methods.