CoRC: Collaborative Partition Optimization for CVRPs

Collaborative Routing Constructors (CoRC), a routing framework by Oguzhan Karaahmetoglu and Hyong Kim submitted to arXiv on 4 July 2026, enables independently solved routing subproblems to exchange customers and vehicles during optimization. The paper evaluates CoRC on AGS benchmark instances and synthetic problems containing up to 200,000 customers and finds that CoRC "consistently constructs feasible routing solutions where competing partition-based methods do not."

What is CoRC and how does it work?

CoRC is a partition-aware routing framework that lets partitioned subproblems trade both customers and vehicles during their local optimization, rather than relying on a fixed partition or a later global re-optimization stage. The opening description in the paper contrasts this collaborative exchange with the standard approach of treating partitions as independent: CoRC interleaves local optimization with customer and vehicle transfers so that demand can be moved to available capacity elsewhere in the fleet.

The paper frames this mechanism as a way to preserve the computational advantages of partitioning while avoiding a common failure mode: unserved customer demand even when fleet capacity exists elsewhere. The authors position collaboration between subproblems as the central operational difference from independent partitioning and from post-routing global fixes.

How did CoRC perform versus other routing methods?

Across AGS benchmark instances and synthetic instances up to 200,000 customers, CoRC outperformed independent partitioning and post-routing global re-optimization on feasibility, and it solved instances that evaluated end-to-end routing frameworks did not finish under the same computational budget. Specifically, the authors report CoRC consistently produced feasible routing solutions across all evaluated partitioning strategies, while competing partition-based methods failed to do so.

The experiments compare CoRC against independent routing, post-routing global re-optimization, and state-of-the-art end-to-end routing frameworks. The paper notes two consistent outcomes: first, partition-based baselines left some customer demand unserved even when capacity existed elsewhere; second, CoRC avoided that failure by allowing transfers of customers and vehicles. The authors also highlight that some end-to-end frameworks did not produce solutions within the same computational budget used in the study, whereas CoRC still produced feasible routes.

Why it matters

Large-scale capacitated vehicle routing is often rendered tractable by partitioning customers, but that step can create hard feasibility gaps. CoRC addresses the gap directly: by making partitions collaborative rather than isolated, it reduces the chance of unserved demand without reverting to a costly full-instance re-optimization. For practitioners, that trade-off matters because it preserves the runtime benefits of partitioning while improving solution feasibility on very large instances, including the paper's synthetic tests up to 200,000 customers.

What to watch

Look for follow-up evaluations that report runtime and service-quality breakdowns across more benchmark sets and real-world fleets, and for whether CoRC appears in peer-reviewed proceedings or as implementation code linked from the arXiv entry. The paper is listed as arXiv:2607.03694 and carries the DOI 10.48550/arXiv.2607.03694; it is 24 pages long with 3 figures and 3 tables, which contain the experimental detail and comparative results.