DS1

DOI10.5281/zenodo.6501500Zenodo6501500MaRDI QIDQ6719971FDOQ6719971

Dataset published at Zenodo repository.

Publication date: 28 April 2022

Copyright license: Creative Commons Attribution 4.0 International

DS1 contains the 1385 estimations of performance of instances of control software observed on physical robots, together with estimations obtained on via the evaluations of the instances on the design model MA, the pseudo-reality model MB, and 1380 pseudo-reality models sampled from the range R. The data is organized in two tables: VerticalDS1.csv is a dataframe composed of 5 columns. Each line corresponds to the execution of an instance of control software in a given context (that is, reality or a simulation model). The column Study contains labels corresponding to the references to the original studies from which we collected the control software and associated real-world performance (see below for an explanation of the labels). The column Method contains the design method used to generate the control software. The column Mission contains the mission that the control software solve. The column Seed corresponds to the initial configuration of the swarm prior to the execution on the physical robots, and the initial configuration and the execution of the simulated swarm. The column Context contains the conditions in which the control software is executed, and the column Score the resulting performance. In the column Context, Reality refers to the physical robots, MA to the design model MA, MB to the pseudo-reality models MB, and the other values are the identifiers of the pseudo-reality models sampled from R (see below for an explanation of the models identifiers). HorizontalDS1.json contains objects that comport 4 name-value pairs. The name InstanceID contains a unique identifier corresponding to one evaluation of an instance of control software on the physical robots (see below for an explanation of the instance identifier). The name Reality contains the real-world performance resulting from the execution of an instance of control software on the physical robots. The name MA contains the performance resulting from the evaluation of the instance on the model MA. The name MB contains the performance resulting from the evaluation of the instance on the model MB. The name xM contains an array of 1380 digits, each corresponding to the performance resulting from the evaluation of the instance on a random model sampled from the range R. Decrypting the study labels: The column Study of Vertical DS1.csv contains identifier for the original study from which we collected data. Here are the different labels and the corresponding references: FraBraBru-etal2015SI = Francesca, G.et al. AutoMoDe-Chocolate: automatic design of control software for robot swarms. Swarm Intell. 9, 125152. (2015). HasLigRudBir2021NATUCOM = Hasselmann, K., Ligot, A., Julian, R. Birattari, M. Empirical assessment and comparison of neuro-evolutionary methods for the automatic off-line design of robot swarms. Nat. Commun. 12, 4345, 10.1038/s41467-021-24642-3 (2021). HasRobBir2018ants = Hasselmann, K., Robert, F. Birattari, M. Automatic design of communication-based behaviors for robot swarms. In Swarm Intelligence ANTS, vol. 11172 of LNCS, 1629 (2018). KucLigBozBir2018ants = Kuckling, J., Ligot, A., Bozhinoski, D. Birattari, M. Behavior trees as a control architecture in the automatic modular design of robot swarms. In Swarm Intelligence ANTS, vol. 11172 of LNCS, 3043 (2018). LigCotGarBir2021 = Ligot, A., Cotorruelo, A., Garone, E. Birattari, M. Towards an empirical practice in off-line fully-automatic design of robot swarms. IEEE Transactions on Evolutionary Computation (2022). LigHasBir2020ants = Ligot, A., Hasselmann, K. Birattari, M. AutoMoDe-Arlequin: neural networks as behavioral modules for the automatic design of probabilistic finite state machines. In Swarm Intelligence ANTS, vol. 12421 of LNCS, 109-122 (2020). SpaKegGarBir2020bnaic-post = Spaey, G., Kegeleirs, M., Garzn Ramos, D. Birattari, M. Comparison of different exploration schemes in the automatic modular design of robot swarms. In Proceedings of the Reference AI ML Conference for Belgium, Netherlands Luxemburg, BNAIC/BENELEARN 2019, vol. 2491 of CEUR Workshop Proceedings (2019). Decrypting the models identifiers: In the column Context of VerticalDS1.csv, pseudo-reality models sampled from R are identified with a label starting with letter M and followed by 12 digits. The digits correspond to the noise injected in the sensors and actuators of the robots, multiplied by 100. Digits 1 and 2 correspond to the noise injected on the wheels, digits 3 and 4 to the noise injected in the proximity sensors, digits 5,6,7 to the loss probability of the range-and-bearing module, digits 8, 9, 10 to the noise injected in the light sensors, digits 11 and 12 to the one injected in the ground sensors. For example, pseudo-reality model MB could be renamed M150509009005 following as the noise parameters for the wheels, proximity, rand-and-bearing, light and ground sensors/actuators are equal to 0.15, 0.05, 0.90, 0.90, and 0.05, respectively. Decrypting the identifier of the evaluation of the instances of control software: In the objects of HorizontalDS1.json, the name InstanceID contains a unique identifier corresponding to one evaluation of an instance of control software on the physical robots. This identifier is composed of 4 elements separated by underscores. The first one is a label corresponding to the study the instance of control software and related performance originate from, and consists in the initials of the authors last names. For example, LHB corresponds to the study label LigHasBir2020ants. The second one is a label corresponding to the mission on which the control software is executed, and consists in the three first letter of the mission name. In the case of the study SpaKegGarBir2020bnaic-post, a B or a U is added to differentiate between the cases in which the mission is taking place in a bounded or unbounded arena, respectively. The third one is a label for the design method used the generate the control software. The fourth and last element is the seed that controls both the initial configuration and execution of the swarm.

This page was built for dataset: DS1