Efficient Feature Mapping Using a Collaborative Team of AUVs

TL;DR

This paper addresses efficient isobath localization using a cooperative team of AUVs by casting level-set estimation as a Bayes-risk minimization problem under a Gaussian-process depth model. The authors propose an informative path planning framework with an augmented receding-horizon scheme and a terminal reward to bound the quality of short-horizon plans, along with an online sparse GP regression to keep computation tractable in the field. The main contributions are a novel objective function based on Bayes' risk and Benefit of Search, plus field experiments that validate performance guarantees under realistic underwater constraints such as slow acoustic communications and limited onboard compute. The results demonstrate that a team of AUVs can achieve efficient, decentralized exploration with near-optimal rewards and robust communication strategies, enabling practical, scalable underwater feature mapping.

Abstract

We present the results of experiments performed using a team of small autonomous underwater vehicles (AUVs) to determine the location of an isobath. The primary contributions of this work are (1) the development of a novel objective function for level set estimation that utilizes a rigorous assessment of uncertainty, and (2) a description of the practical challenges and corresponding solutions needed to implement our approach in the field using a team of AUVs. We combine path planning techniques and an approach to decentralization from prior work that yields theoretical performance guarantees. Experimentation with a team of AUVs provides empirical evidence that the desirable performance guarantees can be preserved in practice even in the presence of limitations that commonly arise in underwater robotics, including slow and intermittent acoustic communications and limited computational resources.

Figures (4)

• Figure 1: 690 AUVs
• Figure 2: Paths and Bayes' risk for experiments accomplished using 3 vehicles. Top two show the prior Bayes' risk while the bottom two panels show the posterior Bayes' risk.
• Figure 3: Results of receding horizon missions conducted with 3 vehicles; the green line is the accumulated reward with receding horizon planning and a terminal reward, the blue dash is the accumulated reward when no terminal reward is used, the black line is the guaranteed lower bound, which is the reward of a naive lawn-mower path, and the red dashed line is the cumulative reward of using receding-horizon planning with terminal reward plus the reward of the naive path.
• Figure 4: The top, middle, and bottom panel show the difference between the posterior Bayes' risk given the measurements available to AUVs 1, 2, and 3, respectively, and the posterior Bayes' risk given all measurements shown in Figure \ref{['fig:3_agent_rh_paths']}. Note that all measurements not obtained by the AUV itself must be received via acoustic communications.

Theorems & Definitions (2)

• Definition 1: Benefit of Search
• Theorem 1