<p>Urban logistics now operates with tighter service windows and greater travel-time variability than the deterministic vehicle routing problem with time windows (VRPTW), was designed to absorb. This article reviews algorithmic approaches for VRPTW under uncertain travel and service times in peer-reviewed journal studies published from 2020 to 2025. The six-year window was fixed before screening to capture the current generation of robust, learning-assisted, and rolling-horizon methods while preserving a closed and reproducible corpus. A PRISMA 2020-guided review was conducted using Scopus, Google Scholar, and a register-based source under explicit eligibility, screening, extraction, and coding rules. Ninety-three primary studies met the final criteria. The review separates the study-level corpus from the algorithm-level corpus and consolidates repeated architectures and parameter-only variants through predefined decision rules, yielding 122 standardised algorithmic instances. The evidence indicates a movement from isolated exact formulations toward hybrid and adaptive architectures that combine robust or stochastic optimisation, advanced metaheuristics, and learning components for prediction, control, or re-optimisation. Comparative findings remain conditional on uncertainty assumptions, instance scale, data availability, and validation design. The review develops an Uncertainty-Resilience Maturity Model (URMM), a contextual selection matrix, and a benchmarking protocol with common uncertainty descriptors and performance indicators. Most evidence still comes from benchmark or simulation studies. Limited operational validation, uneven reporting, and weak transfer tests prevent broad claims of real-world superiority.</p>