A new algorithmic framework called CoRC promises to fix a stubborn blind spot in large-scale delivery logistics software.
When a fleet has tens of thousands of customers to serve, routing software typically slices the problem into smaller chunks and solves each independently. The trouble: those independent solvers sometimes strand customers without a route even when another part of the fleet has spare capacity. CoRC — Collaborative Routing Constructors — addresses this by letting each subproblem trade customers and vehicles with its neighbors during the solve, rather than waiting for a costly global cleanup pass at the end. Researchers tested the framework on benchmark instances and synthetic problems with up to 200,000 customers.
The result matters because "feasibility" — actually serving every customer — is a harder bar than just minimizing total distance. Prior methods frequently failed to reach it at scale; CoRC consistently did, across every partitioning strategy tested, and on problem sizes where end-to-end frameworks timed out entirely. For logistics operators, an infeasible route plan is not a slightly suboptimal one — it is a missed delivery.
The approach does not claim to find the shortest routes, only the ones that work — a modest but genuinely useful promise in a field prone to overselling optimization gains.