A review of Link-level uncertainty in the perception–decision–control pipeline of connected and autonomous vehicles: Generation,evolution,propagation,and amplification

Abstract

The Perception–Decision–Control (PDC) pipeline forms the core architecture enabling environmental perception, path planning, and trajectory execution in Connected and Autonomous Vehicles (CAVs). Uncertainty, as a key factor limiting system reliability and safety, permeates all stages of this pipeline, continuously generation, evolution, propagation, and amplification along the pipeline. This paper focuses on Link-Level Uncertainty management and systematically reviews modeling and mitigation strategies for uncertainty in the PDC pipeline. First, this paper distinguishes between externally induced and internally modeled perception uncertainties. Then, this paper analyzes the evolution of upstream uncertainty within the decision-making module, highlighting modal conflicts and policy ambiguity in trajectory reasoning and intent inference. Subsequently, this paper investigates the downstream propagation of decision uncertainty into the control stage, leading to distorted execution signals and reduced stability. Finally, this paper summarizes mainstream mitigation strategies including perception confidence modeling, robust decision optimization, and fault-tolerant control execution. In contrast to traditional module-specific approaches, future research should emphasize system-level modeling of chain-wise uncertainty evolution, develop cross-layer collaborative representation and optimization mechanisms, and construct end-to-end risk-aware frameworks to ensure safe and robust CAV operation in complex, dynamic environments.