SensingAgent: Advancing Vehicular Sensing Systems for Spatiotemporal Cognitive Intelligence

Abstract

The development of intelligent sensors has garnered widespread attention in autonomous driving. Although they have made significant progress, current vehicular sensing systems are still limited to basic perceptual intelligence and lack deeper cognitive capabilities for comprehensive scene understanding. The emerging LLMs (Large Language Models) and agent technologies provide a promising solution to address these issues. This letter proposes a novel SensingAgent framework for building next-generation vehicular sensing systems toward new AI (Autonomous Intelligence and Agentic Intelligence). It adopts a cloud-edge-end architecture, leveraging multi-agent collaboration to revolutionize the sensing paradigm. Additionally, we introduce DAO (Decentralized Autonomous Organization) into sensing systems and propose a new concept of SAO (Sensor Autonomous Organization). It utilizes smart contracts to ensure trustworthy operations across the sensing industry chain. This letter presents a report on the Distributed/Decentralized Hybrid Workshop on Foundation/Infrastructure Intelligence (DHW-FII), providing new insights into the future of intelligent sensing systems.