Skip to content

Research

One family of problems, ten years.

Ten years on one family of problems, approached from a different angle each time: first as software, then as a benchmark, then as a formulation, and now as something to be solved exactly.

2026 —

Current

Integrated production–routing for radiopharmaceuticals

Supervised by Prof. Dr. Cleber Damião Rocco , School of Applied Sciences (FCA), Unicamp.

The MOPTA 2017 problem couples two decisions that are usually studied apart. A set of heterogeneous production lines makes a radioactive product whose usable dose decays from the moment it is produced; a fleet of vehicles must then deliver it to imaging centres, each of which has a fixed appointment schedule. A batch produced too early is worthless on arrival; a batch produced too late misses the window entirely.

The formulation is a production MIP coupled to a vehicle-routing subproblem with time windows, iterated on a production-time-deadline estimate. The naive coupling diverges — each iteration tightens the deadlines until almost no batch–appointment pair remains feasible — which is itself an interesting failure and the reason a principled decomposition is needed.

A survey across EJOR, Transportation Science and Omega confirms that no published work applies an exact or near-exact method to this problem: the radiopharmaceutical literature uses heuristics, and the exact production-routing literature does not model decay. The intended contribution is an ALNS scheme for the routing subproblem combined with Benders cuts from routing back into production.

Key references

  • Lee et al., EJOR (2014) — nuclear medicine production and delivery
  • Dayarian & Desaulniers, Transportation Science (2019) — branch-price-and-cut for short-lifespan production routing
  • Arda et al., EJOR (2024) — home chemotherapy with stability windows

2016 — 2020

Completed

PostVRP and VRPBench

PostVRP is a vehicle routing variant in which the length of every route is bounded and the objective includes a fairness criterion — routes should distribute workload evenly across carriers, not merely minimise total distance. It came directly from watching how postal delivery actually works: a route that is optimal in aggregate but leaves one carrier with twice the walking is not a solution anyone will accept.

VRPBench generates benchmark instances from real city maps and real delivery data, so that algorithms can be compared on something other than randomly scattered points. The tool was extended to multi-objective instances covering up to 25 vehicles and 30,000 daily deliveries, and published as a chapter in Smart Delivery Systems (Elsevier, 2019).

2015

Completed

TSP applied to postal routing

The starting point: a town’s road network digitised from a map, processed with Bresenham’s line algorithm into a graph of streets and intersections, and solved as a travelling salesman problem. Brute force became infeasible past roughly twenty delivery points, which motivated a branch-and-bound implementation and, eventually, the whole line of work that followed.

The result was delivered as a Java application with a graphical interface, so that a carrier could enter a list of addresses and receive an ordered route. Presented at the XXIII Unicamp Congress of Scientific Initiation, where it received an honourable mention.