Hierarchical Online Learning for Adaptive Sampling of Discrete Species Distributions with an AUV
Jessica Todd, Seth McCammon, Dana Yoerger
AI summary
Problem
Autonomous robots struggle to efficiently sample discrete, sparsely distributed species without prior ecological knowledge, as conventional path planning ignores complex substrate-species correlations and struggles with limited sensing and energy resources.
Approach
The authors combine Gaussian Process regression for substrate mapping with a Log-Gaussian Cox Process and Integrated Nested Laplace Approximation to predict species density online, feeding these predictions into an anytime Monte Carlo Tree Search planner for adaptive sampling.
Key results
- Hierarchical generative model for discrete species distributions over multiple substrates
- Informative path planning algorithm integrating hierarchical inference with adaptive search
- Significant reduction in prediction RMSE over mission duration in synthetic environments
- Up to 70% increase in locating unique coral targets compared to boustrophedon and GP baselines
Why it matters
Provides a scalable, data-driven solution for autonomous marine biodiversity monitoring and conservation without requiring prior ecological priors.
Abstract
Autonomous robots are increasingly being used in the field of scientific exploration and data acquisition. In par- ticular, the use of robotic systems for mapping and sampling of species is becoming widespread in both aerial and underwater domains, however the problem of choosing where to sample is challenging when the phenomena of interest are discrete and sparsely distributed in space or time, such as when mapping a particular benthic species. In this paper we present a hierar- chical online learning framework for reasoning about species distribution in realtime, in order to inform sampling decisions. Drawing inspiration from the Species Distribution Modelling community, a hierarchical probabilistic model is developed using the Integrated Nested Laplace Approximation framework, that enables online inference about expected target hotspots using predicted substrate distributions. Model parameters are learned online to build a prediction over the discrete targets, and the model is integrated into an anytime online planner to enable adaptive path planning. The hierarchical learning approach is demonstrated on simulated synthetic environments and shown to consistently outperform baseline methods such as Gaussian Process regression and boustrophedon coverage approaches, when robot resources are constrained.