Sparse Gaussian process for online seagrass semantic mapping

Guerrero-Font, Eric; Bonin‐Font, Francisco; Martin‐Abadal, Miguel; González-Cid, Yolanda; Oliver-Codina, Gabriel

doi:10.1016/j.eswa.2020.114478

Cited by 7 publications

(7 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The problem of GPs is that they do not scale well, having a time complexity of O(n 3 ). Some works overcome this issue by proposing the use of a sparse version of the GP [24], [28] or local maps fusion using Bayesian Committee Machine (BCM) [16] to decrease time complexity and improve the online execution.…”

Section: B Related Workmentioning

confidence: 99%

“…Furthermore, most relevant strategies, either sampling-based [31] or evolutionary-based [25] do not transfer planning knowledge between consecutive planning iterations. This work is based on the GP modelling described in [28], yet it extends the latter by developing a novel: (a) adaptive visual information gathering (AVIG) framework, (b) decisiontime adaptive replanning (DAR) behavior, and (c) depthfirst Monte Carlo tree search (DF-MCTS) strategy for IPP. In order to deploy such adaptive information gathering in the field we propose a method that joins the advantages of graph-based and sampling-based methods.…”

Section: B Related Workmentioning

confidence: 99%

“…More specifically, it uses the model described in [51], pretrained to discriminate the P. oceanica from the background in underwater images. This model inputs a 480 × 360 pixels resolution RGB image and, as proposed by [28], instead of using the high resolution output of the network (output 0 1 of Fig. 4) the last two transposed convolutional layers can be pruned, reducing the execution time at expenses of a lower output resolution of 30 × 23 pixels (output 0 3 ).…”

Section: B Data Processingmentioning

confidence: 99%

“…The semantic point clouds are generated using the disparity image, and the reference of such point clouds to global coordinates is obtained using the robot pose, as described in [28]. This article has been accepted for publication in a future issue of this journal, but has not been fully edited.…”

Section: B Data Processingmentioning

confidence: 99%

“…The GP, introduced in Section II-A, are useful to build such environment models and provide predictions in non-visited locations, thanks to their uncertainty representation. Previous work presented in [28] provided an extensive study for selecting the type of GP and its configuration for modeling P. oceanica spatial distribution. This aforementioned work pointed out the high performance of using a Sparse Variational GP using MCMC (SGPMC) [53] model with a Matérn function with scaling factor ν = 3 /2, and a beta function to represent the GP likelihood distribution.…”

Section: ) Map Estimationmentioning

confidence: 99%

See 4 more Smart Citations

Adaptive Visual Information Gathering for Autonomous Exploration of Underwater Environments

2021

Self Cite

View full text Add to dashboard Cite

This work presents the development and field testing of a novel adaptive visual information gathering (AVIG) framework for autonomous exploration of benthic environments using AUVs. The objective is to adapt dynamically the robot exploration using the visual information gathered online. This framework is based on a novel decision-time adaptive replanning (DAR) behavior that works together with a sparse Gaussian process (SGP) for environmental modeling and a Convolutional Neural Network (CNN) for semantic image segmentation. The framework is executed in mission time. The SGP uses semantic data obtained from stereo images to probabilistically model the spatial distribution of certain species of seagrass that colonize the sea bottom forming widespread meadows. The uncertainty of the probabilistic model provides a measure of sampling informativeness to the DAR behavior. The DAR behavior has been designed to execute successive informative paths, without stopping, considering the newest information obtained from the SGP. We solve the information path planning (IPP) problem by means of a novel depthfirst (DF) version of the Monte Carlo tree search (MCTS). The DF-MCTS method has been designed to explore the state-space in a depth-first fashion, provide solution paths of a given length in an anytime manner, and reward smooth paths for field realization with non-holonomic robots. The complete framework has been integrated in a ROS environment as a high level layer of the AUV software architecture. A set of simulations and field testing show the effectiveness of the framework to gather data in P. oceanica environments.

show abstract

Section: B Related Workmentioning

confidence: 99%