Perception-Aided AI-RAN for Industrial Physical AI Deployments (Invited Paper)

Abstract

The emerging Physical AI era demands low-latency artificial intelligence (AI) applications and real-time interaction with the physical world, motivating next-generation network designs that go beyond connectivity to provide environmental awareness and intelligent control. The artificial intelligence-radio access network (AI-RAN) concept addresses this by embedding AI capabilities within the radio access network on shared compute infrastructure. However, current AI-RAN architectures lack a dedicated mechanism to bridge raw sensing data and the higher-level environmental understanding needed by AI-native network functions. In this paper, we introduce Environment Intelligence Fabric (EIF), a new workload that bridges physicalworld observations and AI-RAN applications, and propose a perception-aided AI-RAN architecture for industrial deployments. We present an example of the EIF within an Open AI-RAN architecture and then demonstrate an EIF-aided receiver implementation, where a perception-derived channel model complements conventional pilot-based channel estimation, reducing pilot overhead by 50% while maintaining a comparable error vector magnitude (EVM) performance compared to the full-pilot baseline in the evaluated scenario.