Multimodal information-theoretic measures for autonomous exploration

Abstract

Autonomous underwater vehicles (AUVs) are widely used to perform information gathering missions in unseen environments. Given the sheer size of the ocean environment, and the time and energy constraints of an AUV, it is important to consider the potential utility of candidate missions when performing survey planning. In this paper, we utilise a multimodal learning approach to capture the relationship between in-situ visual observations, and shipborne bathymetry (ocean depth) data that are freely available a priori. We then derive information-theoretic measures under this model that predict the amount of visual information gain at an unobserved location based on the bathymetric features. Unlike previous approaches, these measures consider the value of additional visual features, rather than just the habitat labels obtained. Experimental results with a toy dataset and real marine data demonstrate that the approach can be used to predict the true utility of unexplored areas.

Adaptive Visual Information Gathering for Autonomous Exploration of Underwater Environments

“…Information Gathering (IG) algorithms guide such exploration using an information metric which represents the informativeness of the environment variable under study in particular locations, and this metric is used to drive the data recording process towards the more informative spots whilst minimizing a cost, such as the number of measurements, the navigation distance or the mission time. IG algorithms have been used for different types of exploration, for instance; (a) goal-based, where the objective is traveling from an initial location to a goal location with a given cost budget [8], [9], [10], (b) front-based, for traversing a given threshold area [11], [12], (c) frontier-based, usually for indoor environment mapping [13], [14], [15], [16], (d) multimodal, using different data sources [17], [18], (e) multirobot, using multiple robots [19], [20], [21], (f) hotspot-based, to find environmental variable hotspots [22], [23], and (g) coveragebased, for environmental variables dense estimation [24], [25], which is, in fact, the focus of this work.…”

Phytoplankton hotspot prediction with an unsupervised spatial community model

“…Rao et al [12] proposed the use of a neural network to learn a shared representation over multiple sensor modalities for underwater vehicles (imagery and bathymetry). The learned model is then used to identify informationrich locations given exclusively the bathymetric data.…”

An information theoretic approach to sample acquisition and perception in planetary robotics

“…There exists a significant number of exploration planning algorithms capable of navigating and mapping previously unknown environments [6][7][8][9]. However, these algorithms are most often designed to maximize the efficiency of mapping an area without incorporating the possible value of other factors like object analysis (for sampling) and evaluation.…”