A quiet Sunday: no new papers, and a look at GPU routing in practice
The day in optimization
Sunday was quiet, with no new papers or releases to cover. The one item below is a deployment story on fast solvers, which let planners re-run a schedule when conditions change. It isn't dated to this week, so treat it as background, not news.
Theory
Nothing new today.
In practice
Rail maintenance scheduling on a GPU solver. A Slalom write-up on NVIDIA cuOpt describes a rail maintenance scheduling tool built for Kawasaki. According to the write-up, the tool cut the multi-hour weekly job of scheduling repair work to a single task of about ten minutes. Fast solves matter beyond the time saved. If a plan takes ten minutes to produce, a manager can re-plan when a track closes or a crew is unavailable. If it takes hours, the plan tends to stay fixed. These figures come from the consultancy and vendor ecosystem, and no independent check was found.
The same write-up notes that last-mile routing is often put at 40–50% of total logistics cost, which is why routing solvers get so much investment. Vendor savings claims in this area are broad. For example, ORTEC says its optimization typically saves at least 2%, up to 10%, through better vehicle use and fewer empty miles. These are marketing claims and the sources don't date them.
Worth watching
- Whether GPU-based solvers such as cuOpt show independently reported results on real routing and scheduling workloads, not only vendor-linked ones.
Sources: - Down to the last mile: Revolutionizing route optimization with NVIDIA cuOpt - ORTEC optimization capabilities
Try ProvenDecisions free: type, photograph or say an optimization problem; every answer is checked by a solver.