Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Asymptotically-Bounded 3D Frontier Exploration enhanced with Bayesian Information Gain

John Lewis, Meysam Basiri, Pedro U. Lima

Abstract

Robotic exploration in large-scale environments is computationally demanding due to the high overhead of processing extensive frontiers. This article presents an OctoMap-based frontier exploration algorithm with predictable, asymptotically bounded performance. Unlike conventional methods whose complexity scales with environment size, our approach maintains a complexity of $\mathcal{O}(|\mathcal{F}|)$, where $|\mathcal{F}|$ is the number of frontiers. This is achieved through strategic forward and inverse sensor modeling, which enables approximate yet efficient frontier detection and maintenance. To further enhance performance, we integrate a Bayesian regressor to estimate information gain, circumventing the need to explicitly count unknown voxels when prioritizing viewpoints. Simulations show the proposed method is more computationally efficient than the existing OctoMap-based methods and achieves computational efficiency comparable to baselines that are independent of OctoMap. Specifically, the Bayesian-enhanced framework achieves up to a $54\%$ improvement in total exploration time compared to standard deterministic frontier-based baselines across varying spatial scales, while guaranteeing task completion. Real-world experiments confirm the computational bounds as well as the effectiveness of the proposed enhancement.

Asymptotically-Bounded 3D Frontier Exploration enhanced with Bayesian Information Gain

Abstract

Robotic exploration in large-scale environments is computationally demanding due to the high overhead of processing extensive frontiers. This article presents an OctoMap-based frontier exploration algorithm with predictable, asymptotically bounded performance. Unlike conventional methods whose complexity scales with environment size, our approach maintains a complexity of , where is the number of frontiers. This is achieved through strategic forward and inverse sensor modeling, which enables approximate yet efficient frontier detection and maintenance. To further enhance performance, we integrate a Bayesian regressor to estimate information gain, circumventing the need to explicitly count unknown voxels when prioritizing viewpoints. Simulations show the proposed method is more computationally efficient than the existing OctoMap-based methods and achieves computational efficiency comparable to baselines that are independent of OctoMap. Specifically, the Bayesian-enhanced framework achieves up to a improvement in total exploration time compared to standard deterministic frontier-based baselines across varying spatial scales, while guaranteeing task completion. Real-world experiments confirm the computational bounds as well as the effectiveness of the proposed enhancement.

Paper Structure

This paper contains 19 sections, 10 equations, 4 figures, 1 table, 3 algorithms.

Table of Contents

  1. INTRODUCTION
  2. Contributions
  3. METHODOLOGY
  4. SLAM
  5. OctoMap and Frontier Update
  6. OctoMap Update
  7. Forward Detection and Maintenance
  8. Clustering
  9. Best Viewpoint Selection
  10. Path Planning
  11. Enhancement: Bayesian Regressor
  12. EXPERIMENTS & RESULTS
  13. Simulation
  14. Setup
  15. Computation Time
  16. ...and 4 more sections

Figures (4)

  • Figure B1: Proposed exploration framework utilizing frontier approximation to ensure asymptotic viewpoint generation. The dotted lines represent the proposed enhancement assisted by Bayesian Regressor for information gain estimation.
  • Figure B2: Simulation environments used for analysis. Forest baca2021mrs: Unstructured cluttered environment with a tree density of 0.05 trees/$m^2$, Warehouse: Cluttered indoor environment, House Compound: Structured environment, Mars Surface: Minimal feature uneven ground environment, Housing Colony: Structured symmetric environment.
  • Figure C1: Time Averaged Computation across coverage percentage for varying methods for each environment.
  • Figure C2: Real-world setup. From left to right: a) T650-sport UAV equipped with Mid-360 Livox and ASUS NUC 13 Pro. b) Real-world Test Site. c)$3D$ SLAM Map of Real-World Environment with overlaying (Asymptotic, Asymp + Bayes) paths and bounding cuboid. d) Trial averaged mean-std plots of coverage percentage and computation time vs Exploration time.