Agentic Semantic Sensing for Resource-Adaptive AI-RAN

2026-10-06

AI Summary

This study addresses the limitation of existing open-loop, fixed-configuration semantic sensing approaches that cannot adapt to dynamic task evidence by proposing AI-RAN, a closed-loop semantic sensing framework. This framework introduces an agentic semantic sensing paradigm that employs a profile-conditioned causal transformer with key-value cache optimization for efficient state updates. Furthermore, it achieves resource-efficient online sensing through adaptive observation selection and early-exit mechanisms, supported by a theoretical interpretation of information value based on conditional mutual information. Evaluated on the Widar3.0 dataset, the proposed method reduces sensing costs by 25.33% compared to full-sequence high-configuration baselines while retaining 85.79% of the Macro-F1 performance, demonstrating an effective balance between computational efficiency and recognition accuracy.

Abstract

Semantic sensing (SemS) acquires task-relevant information rather than reconstructing complete physical information. Existing SemS formulations typically operate open loop: sensing configurations and observation schedules are fixed before inference and cannot respond to evolving task-level evidence. We propose Agentic SemS, a closed-loop framework for AI-enabled radio access networks (AI-RANs) that controls sensing within a communication-feasible profile set. A profile-conditioned causal Transformer updates the semantic belief from streaming observations, while key-value caching enables efficient state updates across profile changes without repeatedly processing the complete history. A semantic utility network estimates the task-level benefit of acquiring the next observation block under each feasible profile after accounting for sensing cost. The resulting continuation utilities jointly support next-profile selection and semantic early exit, adapting sensing configuration and duration to evolving evidence. The expected semantic gain is further related to conditional mutual information, providing a value-of-information interpretation of continued online sensing. Experiments on Widar3.0 with six emulated sensing profiles show that, in comparison with full-sequence High, the resource-efficient Agentic setting reduces normalized cumulative sensing cost by 25.33% while achieving 85.79% Macro-F1. At the same utility checkpoint, semantic early exit provides a further 12.35% cost reduction over adaptive sensing without early exit, with a 0.97-percentage-point Macro-F1 decrease.

Problem

• Semantic Sensing
• AI-RAN
• Open-loop Sensing
• Resource Adaptation
• Task-level Evidence

Innovation

• Agentic Semantic Sensing
• Profile-conditioned Causal Transformer
• Semantic Utility Network
• Semantic Early Exit
• AI-RAN