ACP Challenge

The Torpedo Scheduling Problem

Torpedo loading hot metal from blast furnace

This problem is a simplification, while still being challenging, of a real life application for steel production. A torpedo as depicted above is used to transport hot metals in a steel production plant. The objective is to schedule optimally all the torpedo moves while satisfying all the operational constraints. The full description of the problem, instances and solution format is available here.

Don't hesitate to comment or ask questions (see contact below) if something is not clear (problem, format of instance, format of solutions, etc).


A set of the 9 instances used for the warm-up challenge is available here as well as a solution checker and an instance generator (in case you want to test on other instances).

Instances of final challenge are now available inline.


Ranking by instance

Since the objective function is lexicographic, the second objective (timeDesulf) will be normalized to have a value always between 0 and 1. In the output of the checker you’ll see:

cost = torpedo + timeDesulf / (upperBoundTorpedo * durDesulf)

The ranking is based on this cost. More precisely it is on the gain with reference to a solution (worst solution) that would use one torpedo for every order at the BF:

gain = (maxNumberTorpedo + 1) - cost

Global and Temporal rankings

Prizes sponsored by n-Side:

  • €250,00 for the winner of the global ranking.
  • €250,00 for the winner of the temporal ranking.

Global ranking

Teams are ranked based on the the sum of normalized gain across all the instances (normalization is done with reference to best gain on each instance). Thus for each instance the contribution of one instance to their global gain is between 0 and 1 (1 if the team obtain the best solution for this instance).

Temporal ranking

Teams are ranked based on number of days where they are ranked first in Global Ranking above (since July 4, 2016).

Recomputation encouraged

We ask to the winner of the challenge to deploy his final winning solver on All the participants are encouraged to do so as well.

At the end of the competition, we'll send a form to collect some information on the techniques used to tackle the problem. We ask all the participants to fill in this form to help us improve subsequent edition of the challenge.

Submission (final and real challenge)

You must register to get your secret TeamsID and then to submit your solutions.


  • September 2015
  • Problem Description

    September 2015

    Problem Description (first set of instances, solution checker...)

  • January 2016
  • Openning of the Warm-up Challenge

    January 15, 2016

    Registration and submission pages are online.

  • April 2016
  • Closing of the Warm-up Challenge

    April 30, 2016

  • May 2016
  • Parameters of the instance generators that will be used for the final set of instances are revealed

    May 1st, 2016

  • June 2016
  • Opening of Real Challenge

    June 1st, 2016

    Final instances revealed.

  • August 2016
  • Closing of Real Challenge

    August 20, 2016

  • September 2016
  • Conference

    September 5-9, 2016



ACP Challenge Chair


  • Cyrille Dejemeppe (UCLouvain)
  • Paul Gaborit (Mines-Albi)
  • Sébastien Mouthuy (n-Side)
  • François-Xavier Mouthuy (n-Side)
  • David Allouche (INRA)
  • Matthias Zytnicki (INRA)
  • Cédric Pralet (ONERA)
  • Nicolas Barnier (ENAC)

Industrial Partners

  • N-SIDE, an optimization company who suggested the problem.
  • ArcelorMittal (Dunkerque, France), a steel production company who was a customer of n-Side on a related project.


New A CP 2016 Google photo album is available

(since September 12, 2016)

Proceedings & Slides

(since September 6, 2016)

CP 2016 Accepted Papers

(since June 7, 2016)