A Multi-Layer Framework for Semantic Perception and Infrastructure Support in Autonomous Driving

Intelligent Transportation Systems (ITS) and Connected and Automated Vehicles (CAVs) represent the next frontier of future mobility. In this context, the perception of the surrounding environment becomes a key enabler for autonomous vehicles, with on-board sensors and Vehicle-to-Everything (V2X) communications playing complementary roles. However, Collective Perception itself must be integrated with intelligent processes to prioritize perceived actors that hold higher relevance for a specific vehicle; this approach is commonly known as semantic perception or semantic communication, which allows the vehicle to assess the potential impact of each object on its immediate driving decisions. This paper presents a four-layer semantic perception framework, supported by an ad-hoc infrastructure to enhance an ego vehicle's motion planning. Experimental results in complex urban scenarios show that the framework provides a median anticipation time of 1.3 seconds for occluded objects, reaching 4.6 seconds at the 95th percentile. Furthermore, the environmental awareness of the assisted vehicle is enhanced by enlarging its Operational Design Domain (ODD). Moreover, the semantic logic reduces communication load by transmitting only 40-50% of perceived objects, with only 20-30% classified as safety-critical, effectively preventing channel congestion.