Begoña González, Diego Rossit, Mariano Frutos, Máximo Méndez (2025), Annals of Operations Research (vol. 350, pp. 979–1015) — ARXIV-2504.10648
Key takeaways
- A mixed binary/integer-encoded genetic algorithm solves the joint bin-sizing and weekly-routing problem for municipal waste collection. On small instances it comes close to Gurobi’s MILP/MIQP solutions within a similar time budget, and it stays feasible on a real 163-point instance from Bahía Blanca, where the exact models are intractable.
Seeded from the abstract and a skim, before the reading in NOTE-015. What follows is what the work says about itself.
Two waste-collection decisions are usually optimized separately, because each is already hard: how much bin capacity to place at each collection point, and how to design and schedule the collection routes. This paper couples them in one model. It gives two exact formulations for this model and a genetic algorithm whose chromosome mixes binary and integer genes. The authors test several crossover and mutation operators to find a good configuration. The GA matches the exact solvers on small instances and still finds feasible solutions on large, realistic instances in reasonable time.
Standing in the record
Filed by the reading-time triage of 2026-09-25: 1,870 seconds of active reading over 5 sessions in the papers-feed tracker. It was filed Deferred, unread. NOTE-015 is the close reading of 2026-09-25, and it placed the work: Active — worth reading narrowly, for its formulation (Eqs. 1–2m, with the linearisation in 3a–5) and its public Bahía Blanca instances. The headline “GA matches exact solvers” should not be cited as a quality claim: the exact comparators are unproven incumbents with ~30% gaps, the GA was tuned on one of its five test instances, and the large instances have no reference at all.
When seeded, before the reading in NOTE-015, the entry gave a priority for a deeper reading: low — An applied operations-research paper whose contribution the skim captures. The anthology has little use for it, and t = 1870 s suggests the owner has already read a good part of it.
What the seeding asked a deeper reading to check (NOTE-015 is that reading):
- The claim that the GA is “competitive” rests on five 12-point instances with an 8-hour budget for the exact solvers. Check whether the exact solvers’ optimality gaps were closed and whether 30 GA runs × evaluations make a fair comparison.
- The link to ML is thin: an evolutionary metaheuristic for combinatorial optimization, with no learning component.
Access when seeded: The arXiv abs page and the full 35-page PDF (v1, 14 Apr 2025) were both reachable. I confirmed the journal DOI and venue through Crossref (published 2025-05-18). I read the contributions (§2.1), the section heads, the results (§6, Tables 4–6) and the conclusion (§7).