Bauschmann, NathalieNathalieBauschmann2024-10-212024-10-212024-09-1835. Forum Bauinformatik, fbi 2024: 107-114https://hdl.handle.net/11420/49624Autonomous underwater vehicles (AUVs) can serve as efficient and cost-effective mobile sensor nodes for monitoring of underwater environments and construction sites such as marinas or nuclear storage ponds. In field exploration, the information collected by the AUV is used to construct a probabilistic field belief model of the environmental field. This can then be used to integrate an information metric into the path planning step. In this paper, a Gaussian Process-based field belief for a scalar, 2D underwater environmental field is investigated. The main contribution is the extensive experimental validation using real measurements. For testing purposes, a radiation field is mimicked using electro-magnetic carrier signals emitted by spatially distributed active beacons. Leveraging the attenuation of these signals underwater, the AUV measures the received signal strength. In literature so far, field exploration frameworks in the underwater domain are typically validated in experiments by simulating the field itself and the resulting measurements. While sensor noise can be integrated relatively easily, this still constitutes an over-simplification due to model imperfections. Moreover, this approach neglects other effects resulting from e.g. dynamic motion as the AUV interacts with the environment or field anomalies due to obstacles. Hence, an experimental investigation under realistic conditions is expected to deliver insights.enhttps://creativecommons.org/licenses/by/4.0/Gaussian Process regressionProbabilistic field modelingunderwater roboticsNatural Sciences and Mathematics::500: ScienceTechnology::621: Applied Physics::621.8: Machine EngineeringNatural Sciences and Mathematics::519: Applied Mathematics, ProbabilitiesConference Paper10.15480/882.1353110.15480/882.13531Conference Paper