Online Informative Path Planning for Active Information Gathering of a 3D Surface

Hai Zhu, Jen Jen Chung, Nicholas R. J. Lawrance, Roland Siegwart, Javier Alonso-Mora

TL;DR

This paper introduces an online path planning method for UAVs to efficiently gather information from 3D surfaces. By using manifold Gaussian processes to model spatial correlations in the data and planning in a receding horizon, the approach significantly improves information gathering efficiency compared to traditional coverage methods.

What Problem Does This Paper Address?

Inspecting 3D structures with UAVs is becoming increasingly common for tasks like bridge maintenance or aircraft inspection. Existing methods often plan paths offline to ensure complete surface coverage, but they make a crucial assumption: that the information (e.g., temperature, crack presence) is uniformly distributed. This ignores the reality that information can be spatially correlated, meaning knowing the value at one point gives you information about nearby points. Furthermore, offline planning can't adapt to new information gathered during the mission. This paper tackles these limitations by proposing an online, adaptive path planning strategy that leverages spatial correlations to gather information more efficiently within a given time budget.

Methods

The core of the proposed method lies in two key components:

  • Manifold Gaussian Processes (mGPs) for Surface Mapping: Instead of treating the surface as a flat 2D plane or a simple 3D space, the authors use mGPs with geodesic kernel functions. This allows them to model the information field (e.g., temperature) on complex 3D surfaces while explicitly capturing spatial correlations. The geodesic kernel specifically accounts for distances along the surface itself, not just Euclidean distances in 3D space.
  • Online Informative Path Planning (IPP) with Receding Horizon: The planning is done online, meaning the UAV makes decisions based on its current knowledge and plans a short segment of its path (receding horizon). This allows it to adapt to new measurements. The approach uses a greedy search to identify promising viewpoints from a library, followed by continuous optimization of a polynomial trajectory. The objective is to maximize information gain (measured by the reduction in map uncertainty) while respecting the UAV's dynamic constraints and avoiding collisions. Bayesian fusion is used to efficiently update the mGP map with new sensor data.

Datasets

The paper does not introduce new datasets. Instead, it uses simulated data for evaluation:

  • Cylinder Storage Tank: A 3D model of a cylinder storage tank was created in SolidWorks, and simulated temperature distributions were generated as ground truth.
  • Boeing-747 Airplane: A coarse 3D model of a Boeing-747 airplane was used to simulate temperature distribution for a more complex inspection scenario.

Key Insights and Ideas

  • Spatial Correlations Matter: The most significant insight is that explicitly modeling spatial correlations in the information field (using mGPs) dramatically improves information gathering efficiency. Ignoring these correlations leads to less effective data collection.
  • Online Adaptation is Crucial: Planning paths offline is suboptimal because it can't leverage new information gathered during flight. An online, receding horizon approach allows the UAV to dynamically adjust its path to gather the most valuable data.
  • Information Gain vs. Coverage: The paper highlights that simply covering the entire surface isn't always the most efficient way to gain information. Informative path planning prioritizes areas that will yield the most significant reduction in uncertainty, leading to faster and more accurate mapping.
  • Geodesic Kernels for Surfaces: Using geodesic kernels within mGPs is key to accurately modeling correlations on non-planar surfaces, which is a significant step beyond standard GP applications.

Key Results

  • Improved Uncertainty Reduction: The proposed IPP method significantly reduced map covariance uncertainty compared to traditional coverage path planning and random exploration strategies. For instance, in the cylinder inspection, the informative path achieved much faster uncertainty reduction, particularly in the early stages.
  • Reduced Inspection Error: Similarly, the RMSE of the inspection results with respect to the ground truth was lower with the proposed IPP method, indicating more accurate reconstructions.
  • Impact of Correlation Modeling: When comparing different map initialization methods, using mGPs (which encode spatial correlations) resulted in significantly better performance than using an identity matrix (no correlation) or a random SPD matrix (random correlations).
  • Successful Complex Inspection: The method was successfully demonstrated on a complex Boeing-747 airplane model, showing continuous reduction in map uncertainty and error throughout the mission.

Limitations and Future Work

  • Local Minima: The authors acknowledge that the local nature of the receding horizon planning might lead to the UAV getting trapped in local minima.
  • Future Work:
    • Global Path Planning Integration: Combining the local IPP with a global path planner could help overcome the local minima issue.
  • Dynamic Information Fields: Extending the approach to handle information fields that change over time.
  • Multi-Robot Information Gathering: Investigating how multiple UAVs can collaborate for even more efficient data acquisition.
  • Occlusion Handling: While simplified for convex surfaces, more robust occlusion checking for complex geometries is a potential area for improvement.

Summary updated

Abstract

This paper presents an online informative path planning approach for active information gathering on three-dimensional surfaces using aerial robots. Most existing works on surface inspection focus on planning a path offline that can provide full coverage of the surface, which inherently assumes the surface information is uniformly distributed hence ignoring potential spatial correlations of the information field. In this paper, we utilize manifold Gaussian processes (mGPs) with geodesic kernel functions for mapping surface information fields and plan informative paths online in a receding horizon manner. Our approach actively plans information-gathering paths based on recent observations that respect dynamic constraints of the vehicle and a total flight time budget. We provide planning results for simulated temperature modeling for simple and complex 3D surface geometries (a cylinder and an aircraft model). We demonstrate that our informative planning method outperforms traditional approaches such as 3D coverage planning and random exploration, both in reconstruction error and information-theoretic metrics. We also show that by taking spatial correlations of the information field into planning using mGPs, the information gathering efficiency is significantly improved.
BibTeX
@inproceedings{zhu2021online,
  author = {Hai Zhu and Jen Jen Chung and Nicholas R. J. Lawrance and Roland Siegwart and Javier Alonso-Mora},
  title = {{Online Informative Path Planning for Active Information Gathering of a 3D Surface}},
  booktitle = {IEEE International Conference on Robotics and Automation},
  year = {2021},
  url = {https://openreview.net/forum?id=Vt_AN_sSimn},
}
Figure