Multi-path planning based on a NSGA-II for a fleet of robots to work on agricultural tasks
- Publikations-Art
- Kongressbeitrag
- Autoren
- Jesus Conesa-Muñoz and Angela Ribeiro and Dionisio Andújar and Fernández-Quintanilla and Jose Dorado
- Erscheinungsjahr
- 2012
- Veröffentlicht in
- IEEE Congress on Evolutionary Computation (CEC)
- Herausgeber
- IEEE Press. IEEE Computational Intelligence Society
- ISBN / ISSN / eISSN
- 978-1-4673-1510-4
- DOI
- 10.1109/CEC.2012.6256629
- Seite (von - bis)
- 2236--2243
- Tagungsname
- World Congress on Computational Intelligence (WCCI 2012)
- Tagungsort
- Brisbane, Australia
- Tagungsdatum
- June 10--15
In many situations, using multiple robots in the same environment is a good strategy to handle tasks that are too complex or even too expensive for a single robot. One of these situations is the automation of tasks in the agricultural environment. In this context, one of the main problems consists of determining the best routes (multi-path plan) for the robots to minimise cost, while ensuring a fully completed treatment, i.e., the whole field is covered. The cost can be expressed by a function that considers the most relevant features of each robot in the fleet, for example, in a spray weed treatment, the tank capacity, the number of turns required or the time spent in the whole treatment. This multi-path planning problem can be expressed as a bi-objective problem. In particular, in this paper, two different objectives are taken into account: the cost in time and the cost in money. This formulation allows the analysis of situations in which it is important to distribute the robots to reduce the
time of the treatment independently of the money spent and of situations where it is important to reduce the spent money independently of the time consumed. A Non-dominated Sorting Genetic Algorithm II (NSGA-II) is proposed for solving the multi-objective problem. The proposed approach has proven to offer good results in multiple situations dealing with different fields and robots with diverse features. Moreover, the results obtained show that it is possible to determine solutions very close to the optimum of each objective, even simultaneously.