This work presents an approach for embodied agents that have to learnmodels from the least amount of prior knowledge, solely based on knowing whichactions can be performed and observing the state. Instead of relying on (oftenblack-box) quantitative models, a qualitative forward model is learned that findsthe relations among the variables, the contextual relations, and the qualitative influence.We assume qualitative determinism and monotonicity, assumptions motivatedby human learning. A learning and exploitation algorithm is designed anddemonstrated on a robot with a gripper. The robot can grab an object and moveit to another location, without predefined knowledge of how to move, grab ordisplace objects.
Lemeire, J, Wouters, N, Van Cleemput, M & Heirman, A 2023, Learning One Abstract Bit at a Time Through Self-invented Experiments Encoded as Neural Networks. in CL Buckley, D Cialfi, P Lanillos, M Ramstead, T Verbelen, M Ramstead, N Sajid, H Shimazaki & M Wisse (eds), Active Inference, Iwai 2023. vol. 1915, Communications In Computer And Information Science, Springer Nature, pp. 254-274, the 4th International Workshop on Active Inference (IWAI 2023), 13-15 September 2023, Ghent, Belgium, Ghent, Belgium, 13/09/23. https://doi.org/10.1007/978-3-031-47958-8_16
Lemeire, J., Wouters, N., Van Cleemput, M., & Heirman, A. (2023). Learning One Abstract Bit at a Time Through Self-invented Experiments Encoded as Neural Networks. In C. L. Buckley, D. Cialfi, P. Lanillos, M. Ramstead, T. Verbelen, M. Ramstead, N. Sajid, H. Shimazaki, & M. Wisse (Eds.), Active Inference, Iwai 2023 (Vol. 1915, pp. 254-274). (Communications In Computer And Information Science). Springer Nature. https://doi.org/10.1007/978-3-031-47958-8_16
@inproceedings{31f87e82fff84c29b10335193c60f88a,
title = "Learning One Abstract Bit at a Time Through Self-invented Experiments Encoded as Neural Networks",
abstract = "This work presents an approach for embodied agents that have to learnmodels from the least amount of prior knowledge, solely based on knowing whichactions can be performed and observing the state. Instead of relying on (oftenblack-box) quantitative models, a qualitative forward model is learned that findsthe relations among the variables, the contextual relations, and the qualitative influence.We assume qualitative determinism and monotonicity, assumptions motivatedby human learning. A learning and exploitation algorithm is designed anddemonstrated on a robot with a gripper. The robot can grab an object and moveit to another location, without predefined knowledge of how to move, grab ordisplace objects.",
keywords = "Autonomous robots, Developmental learning, Open-ended learning, Qualitative Models",
author = "Jan Lemeire and Nick Wouters and {Van Cleemput}, Marco and Aron Heirman",
year = "2023",
doi = "10.1007/978-3-031-47958-8_16",
language = "English",
isbn = "978-3-031-47957-1",
volume = "1915",
series = "Communications In Computer And Information Science",
publisher = "Springer Nature",
pages = "254--274",
editor = "Buckley, {Christopher L.} and Daniela Cialfi and Pablo Lanillos and Maxwell Ramstead and Tim Verbelen and Maxwell Ramstead and Noor Sajid and Hideaki Shimazaki and Martijn Wisse",
booktitle = "Active Inference, Iwai 2023",
note = "the 4th International Workshop on Active Inference (IWAI 2023), 13-15 September 2023, Ghent, Belgium, IWAI ; Conference date: 13-09-2023 Through 15-09-2023",
url = "https://iwaiworkshop.github.io/",
}