Toward a universal perception layer: A survey on sensor-agnostic advanced driver assistance systems

Modern vehicle portfolios of automakers span multiple platforms, trim levels, and region-specific configurations, each imposing different hardware constraints and functional requirements. For environment perception, this variation is difficult to manage because current automotive perception stacks are typically designed around fixed sensor layouts, calibration assumptions, and compute environments. As sensor hardware is updated, repositioned, omitted, or degraded over a vehicle program’s lifetime, these tightly coupled designs can require substantial redevelopment and revalidation effort. This survey formalizes the concept of a universal perception layer, an architectural abstraction that decouples environment perception from the underlying sensor hardware and provides a stable, modality-independent interface to downstream planning and control modules. We concretize this concept through three architectural stages: feature extraction, fusion, and representation. These stages are related to a sensor-variation taxonomy that distinguishes sensor-instance variation, sensor-suite variation, and sensor-state variation. Through this lens, the survey synthesizes relevant methods, situates them within the broader sensor landscape, and examines the data resources needed to evaluate sensor-agnostic perception. ... mehrDespite substantial progress, important challenges remain, including architectural limitations, missing benchmarks, and the need for training data that spans diverse sensor configurations. We highlight these open problems and outline future research directions toward more flexible, robust, and scalable perception stacks for next-generation intelligent vehicles.