Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Efficient Non-Myopic Layered Bayesian Optimization For Large-Scale Bathymetric Informative Path Planning

Alexander Kiessling, Ignacio Torroba, Chelsea Rose Sidrane, Ivan Stenius, Jana Tumova, John Folkesson

TL;DR

This work presents a 2-layered BO IPP method that performs non-myopic, online planning in a tree search fashion over large Stochastic Variational GP maps, while respecting the AUV dynamical constraints and accounting for localization uncertainty.

Abstract

Informative path planning (IPP) applied to bathymetric mapping allows AUVs to focus on feature-rich areas to quickly reduce uncertainty and increase mapping efficiency. Existing methods based on Bayesian optimization (BO) over Gaussian Process (GP) maps work well on small scenarios but they are short-sighted and computationally heavy when mapping larger areas, hindering deployment in real applications. To overcome this, we present a 2-layered BO IPP method that performs non-myopic, real-time planning in a tree search fashion over large Stochastic Variational GP maps, while respecting the AUV motion constraints and accounting for localization uncertainty. Our framework outperforms the standard industrial lawn-mowing pattern and a myopic baseline in a set of hardware in the loop (HIL) experiments in an embedded platform over real bathymetry.

Efficient Non-Myopic Layered Bayesian Optimization For Large-Scale Bathymetric Informative Path Planning

TL;DR

This work presents a 2-layered BO IPP method that performs non-myopic, online planning in a tree search fashion over large Stochastic Variational GP maps, while respecting the AUV dynamical constraints and accounting for localization uncertainty.

Abstract

Informative path planning (IPP) applied to bathymetric mapping allows AUVs to focus on feature-rich areas to quickly reduce uncertainty and increase mapping efficiency. Existing methods based on Bayesian optimization (BO) over Gaussian Process (GP) maps work well on small scenarios but they are short-sighted and computationally heavy when mapping larger areas, hindering deployment in real applications. To overcome this, we present a 2-layered BO IPP method that performs non-myopic, real-time planning in a tree search fashion over large Stochastic Variational GP maps, while respecting the AUV motion constraints and accounting for localization uncertainty. Our framework outperforms the standard industrial lawn-mowing pattern and a myopic baseline in a set of hardware in the loop (HIL) experiments in an embedded platform over real bathymetry.

Paper Structure

This paper contains 17 sections, 6 equations, 4 figures, 2 tables.

Table of Contents

  1. INTRODUCTION
  2. RELATED WORK
  3. METHODOLOGY
  4. Scalable Bathymetry Regression Under Uncertainty
  5. Layer 1: BO-based Next Best Viewpoint
  6. Non-Myopic Tree Search
  7. Batch BO for Efficient Tree Search
  8. Layer 2: Constrained Path Optimization
  9. Fail-safe Behaviors
  10. EXPERIMENTAL SETUP
  11. Simulation Environment
  12. Experiments Configuration
  13. Performance evaluation
  14. Results
  15. Bathymetric surveying
  16. ...and 2 more sections

Figures (4)

  • Figure 1: Our 2-layer IPP: the latest AUV pose and the MBES data collected so far are used to regress online a SVGP-UI model of the bathymetry (layer 1) over which a non-myopic BO stage selects the optimal next viewpoint. A feasible path is then selected on a second BO stage (layer 2) by maximizing the expected information collected.
  • Figure 2: HIL setup in a Jetson Orin: simulated MBES data is generated from real bathymetry with an AUV equipped with a sonar model. Our IPP receives real time odometry and MBES data from the simulator and outputs paths.
  • Figure 3: RMS consistency errors over distance of the SVGP-UI maps posteriors against the ground truth bathymetry. The SVGP-UIs regressed with our method converge faster on average (blue) than the lawn-mowing pattern ones (purple) and myopic method (red).
  • Figure 4: Time lapse of a survey run over area 2 with the AUV path (black) over the SVGP-UI maps posteriors at three time steps. Our IPP has fully mapped the area while the LM has completed $60\%$ of the path. Original bathymetry on the left.