Jean-Christophe Laneri & Nicolas Renaux École Supérieure d’Informatique Electronique Automatique Ave Maurice Thorez, Ivry sur Seine, France

Advisor : Sophie Maucorps Emails : {laneri, maucorps, renaux}@esiea.fr

About $ 10bn worth of art treasures are stolen and traded around the world every year. We have developed a parallel genetic algorithm (RLM) to place laser beams in a museum gallery. RLM performs a multi-objective optimization: it maximizes the security of the gallery and minimizes the number of laser beams.

Expert

RLM

Parameters

Top Configurations

Decision

Gallery Constraints : dimensions, doors, windows, treasures locations & dimensions, … RLM Parameters : quantity of individuals & generations, number of robbers who test each configuration, the maximal distance they can traverse, …

Fitness

Crossing-Over Individuals are mated at random. Then, each pair undergoes crossing-over as follows: four new individuals are created by swapping half of the ancestors’ chromosomes.

Mutation

Best Individuals

Standard Individuals

Best and Standard Individuals converge on an optimal adaptation

Example of four individuals Ind. 4 Ind. 3

Ind. 1 Ind. 2

Centralized Implementation

Selection

Parallel Implementation

Example of robber attack

On the left, a “human” configuration. RLM one (on the right) do better : 40 % in terms of security, and 15 % in terms of laser beams number.

At each generation, It allows one individual per cent to have one more chromosome.

RLM can easily be adapted to solve the Art Gallery Problem in three dimensions by using cameras instead of laser beams. In this case, cones would replace the segments.

Thanks : Laurent Beaudoin, Imelda Chantomaud, Stephane Duval & Robert Erra for helping us during the last six months Michel Albuisson and the Remote Sensing Group of Ecole des Mines de Paris for printing this poster ACM Student Research Competition, SIGCSE conference, Norfolk, Virginia, March 2004