Abstract:Robotic Information Gathering (RIG) is a foundational research topic that answers how a robot (team) collects informative data to efficiently build an accurate model of an unknown target function under robot embodiment constraints. RIG has many applications, including but not limited to autonomous exploration and mapping, 3D reconstruction or inspection, search and rescue, and environmental monitoring. A RIG system relies on a probabilistic model's prediction uncertainty to identify critical areas for informative data collection. Gaussian Processes (GPs) with stationary kernels have been widely adopted for spatial modeling. However, real-world spatial data is typically non-stationary -- different locations do not have the same degree of variability. As a result, the prediction uncertainty does not accurately reveal prediction error, limiting the success of RIG algorithms. We propose a family of non-stationary kernels named Attentive Kernel (AK), which is simple, robust, and can extend any existing kernel to a non-stationary one. We evaluate the new kernel in elevation mapping tasks, where AK provides better accuracy and uncertainty quantification over the commonly used stationary kernels and the leading non-stationary kernels. The improved uncertainty quantification guides the downstream informative planner to collect more valuable data around the high-error area, further increasing prediction accuracy. A field experiment demonstrates that the proposed method can guide an Autonomous Surface Vehicle (ASV) to prioritize data collection in locations with significant spatial variations, enabling the model to characterize salient environmental features.

Vision-Based Satellite Recognition and Pose Estimation Using Gaussian Process Regression

A spatial Gaussian process method for hyperspectral remote sensing imagery classification

GP-SLAM plus : real-time 3D lidar SLAM based on improved regionalized Gaussian process map reconstruction

Gaussian process and its applications in hyperspectral image classification

Gaussian Process Approach to Remote Sensing Image Classification

Gaussian Particle Filter Based Pose and Motion Estimation

From Satellite to Ground: Satellite Assisted Visual Localization with Cross-view Semantic Matching

A Novel Hyperspectral Remote Sensing Images Classification Using Gaussian Processes with Conditional Random Fields

Three-Dimensional Measurement Method of Four-View Stereo Vision Based on Gaussian Process Regression

Pose Estimation for Cross-Domain Non-Cooperative Spacecraft Based on Spatial-Aware Keypoints Regression

Human motion tracking by temporal-spatial local gaussian process experts

UAV Pose Estimation using Cross-view Geolocalization with Satellite Imagery

Object Gaussian for Monocular 6D Pose Estimation from Sparse Views

Adaptive Robotic Information Gathering via Non-Stationary Gaussian Processes

A Monocular Structured Light Vision Method for Pose Determination of Large Non-Cooperative Satellites

A Gaussian Process Classification and Target Recognition Algorithm for SAR Images

A Vision-Based Pose Estimation of a Non-Cooperative Target Based on a Self-Supervised Transformer Network

GPV-Pose: Category-level Object Pose Estimation Via Geometry-guided Point-wise Voting

RRVPE: A Robust and Real-Time Visual-Inertial-GNSS Pose Estimator for Aerial Robot Navigation

UAV Pose Estimation in GNSS-Denied Environment Assisted by Satellite Imagery Deep Learning Features

Manifold Gaussian Processes for Regression