Answers are grounded in this paper and cite their sources.
Onboard autonomous decision-making is a pivotal capability for Autonomous Underwater Vehicles (AUVs) operating in remote, communication-limited, and hazardous environments. This paper presents a state-of-the-art survey of decision-making architectures, examining rule-based and finite-state machines (FSMs), behavior-based architectures, deliberative model-based reasoning, and hybrid architectures, together with the Artificial Intelligence (AI) contributions increasingly embedded within each. Reviewing real-world implementations across AUV platforms, the analysis finds hybrid architectures to be the most widely adopted arrangement in practice, while showing that each architecture remains appropriate to a different mission profile and that no architecture dominates. Five implementation gaps persist: mission-level risk management, runtime assurance under embedded power and compute constraints, integrated mission-level reasoning, sim-to-real transfer, and explainability for learned modules. Alongside these, the evidence base itself is deficient, as the literature lacks comparative benchmarks under shared mission conditions and reports availability, maintenance, and modeling effort too sparsely to substitute for them.
No takes yet. Share an insight, caveat, or question.
Abreu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: