The paper by Paul Shaw from ILOG S.A. explores the application of Large Neighbourhood Search (LNS) to solve Vehicle Routing Problems (VRPs). LNS is a local search method that explores a large neighborhood of the current solution by selecting and removing a set of "related" customer visits from the planned routes, and then re-inserting these visits using a constraint-based tree search. Unlike other methods, LNS uses Limited Discrepancy Search (LDS) during the tree search to re-insert visits, ensuring cost and legality. The authors analyze the performance of LNS on benchmark VRP problems, demonstrating that it produces competitive results with Operations Research meta-heuristic methods. This indicates that constraint-based technology can effectively solve VRPs, leveraging the strengths of both exploration and propagation. The paper is structured into four sections: an introduction, a detailed description of LNS, computational experiments on benchmark problems, and a conclusion. The introduction highlights the challenges and benefits of applying LNS to VRPs, while the second section delves into the specifics of LNS, including the process of continual relaxation and re-optimization. The third section presents experimental results, and the fourth section concludes the study.The paper by Paul Shaw from ILOG S.A. explores the application of Large Neighbourhood Search (LNS) to solve Vehicle Routing Problems (VRPs). LNS is a local search method that explores a large neighborhood of the current solution by selecting and removing a set of "related" customer visits from the planned routes, and then re-inserting these visits using a constraint-based tree search. Unlike other methods, LNS uses Limited Discrepancy Search (LDS) during the tree search to re-insert visits, ensuring cost and legality. The authors analyze the performance of LNS on benchmark VRP problems, demonstrating that it produces competitive results with Operations Research meta-heuristic methods. This indicates that constraint-based technology can effectively solve VRPs, leveraging the strengths of both exploration and propagation. The paper is structured into four sections: an introduction, a detailed description of LNS, computational experiments on benchmark problems, and a conclusion. The introduction highlights the challenges and benefits of applying LNS to VRPs, while the second section delves into the specifics of LNS, including the process of continual relaxation and re-optimization. The third section presents experimental results, and the fourth section concludes the study.