Learning-Based, Bio-Inspired, and Hybrid Methods for Vehicle Routing: A Critical Review of Experimental Evidence

Background: Learning-based, bio-inspired, and hybrid methods report routing improvements, but differences in experimental design complicate comparison. This critical review examines the evidence supporting these claims. Methods: Structured screening and critical synthesis covered 240 journal studies from 2022 to 2026: 43 learning-based, 135 bio-inspired and evolutionary, and 62 hybrid. Full-text assessment using category-specific thresholds yielded 1243 primary decisions, complemented by resource-stage and bio-inspired evaluation analyses. Results: Learning generalization and controlled ablation were Supported in 14/43 and 17/43 studies; none met complete joint resource-accounting or live operational-validation thresholds. Bio-inspired studies reported repeated runs in 104/135 cases and public code with inputs in 9/135; complementary feasibility, fairness, and generalization thresholds were Supported in 10/135, 4/135, and 2/135. Among 56 hybrids with comparative routing outcomes, 39 met neither the matched-budget nor the integration-cost threshold. Conclusions: This review contributes a category-specific framework linking routing improvements to feasibility, experimental controls, computational resources, and validation scope. Under its operational thresholds, these reporting frequencies characterize evidence rather than overall study quality. The framework helps researchers and logistics practitioners distinguish Supported claims from those requiring additional evidence, without pooling incompatible performance results or ranking methodological families.