[Paper Review] Multiresolution Mapping and Informative Path Planning for UAV-based Terrain Monitoring

Marija Popović, Teresa Vidal‐Calleja+4 more|arXiv (Cornell University)|Mar 8, 2017

Robotics and Sensor-Based Localization17 references4 citations

TL;DR

This paper proposes a multiresolution Gaussian Process-based mapping and informative path planning framework for UAVs to monitor agricultural biomass efficiently. By fusing variable-resolution aerial imagery using a GP prior with altitude-dependent sensor models, it enables constant-time map updates and optimizes 3D trajectories via CMA-ES, reducing mean error by up to 45% compared to lawn mower coverage in simulations and demonstrating real-time performance on a multirotor.

ABSTRACT

Unmanned aerial vehicles (UAVs) can offer timely and cost-effective delivery of high-quality sensing data. How- ever, deciding when and where to take measurements in complex environments remains an open challenge. To address this issue, we introduce a new multiresolution mapping approach for informative path planning in terrain monitoring using UAVs. Our strategy exploits the spatial correlation encoded in a Gaussian Process model as a prior for Bayesian data fusion with probabilistic sensors. This allows us to incorporate altitude-dependent sensor models for aerial imaging and perform constant-time measurement updates. The resulting maps are used to plan information-rich trajectories in continuous 3-D space through a combination of grid search and evolutionary optimization. We evaluate our framework on the application of agricultural biomass monitoring. Extensive simulations show that our planner performs better than existing methods, with mean error reductions of up to 45% compared to traditional "lawnmower" coverage. We demonstrate proof of concept using a multirotor to map color in different environments.

Motivation & Objective

• To improve the efficiency of UAV-based terrain monitoring by optimizing data collection in complex, spatially correlated environments.
• To address the challenge of fusing multiresolution aerial imagery with altitude-dependent uncertainty into a single probabilistic map.
• To enable real-time, information-rich trajectory planning that balances resolution, field of view, and resource constraints.
• To validate the framework in simulations and real-world indoor tests using a multirotor UAV.
• To demonstrate applicability to scalar field mapping beyond agriculture, such as gas concentration or elevation.

Proposed method

• Uses Gaussian Processes (GPs) as a prior to encode spatial correlations in continuous scalar fields like biomass.
• Applies Bayesian fusion with altitude-dependent sensor models to enable constant-time map updates using sequential measurements.
• Employs a 3-D grid search to initialize candidate trajectories, followed by CMA-ES evolutionary optimization to maximize information gain.
• Introduces a parametrization for realistic sensor dynamics and adaptive focusing on high-infestation (high-uncertainty) regions.
• Utilizes a depth camera for real-time color saturation mapping in indoor experiments, simulating vegetation detection.
• Employs a Matérn kernel for GP training and defines utility based on uncertainty reduction and thresholded biomass levels.

Experimental results

Research questions

• RQ1 How can multiresolution aerial imagery with varying resolution and uncertainty be fused efficiently into a single probabilistic map?
• RQ2 Can a GP-based prior improve the accuracy and efficiency of UAV-based terrain monitoring in complex, spatially correlated environments?
• RQ3 How does evolutionary optimization of 3D trajectories compare to traditional coverage patterns in terms of information gain and error reduction?
• RQ4 To what extent can the framework operate in real time with limited computational and battery resources?
• RQ5 Can the approach be generalized to other scalar field monitoring tasks beyond agriculture?

Key findings

• The proposed planner reduced mean error by up to 45% compared to traditional lawn mower coverage in simulations.
• CMA-ES outperformed other optimization methods, achieving the lowest uncertainty and error metrics across 30 trials.
• Real-time experiments on a DJI Matrice 100 demonstrated successful mapping of painted green sheets using a depth camera and Vicon-based state estimation.
• Uncertainty decreased significantly over time in all test scenarios, validating the framework's ability to focus on high-information regions.
• The system maintained real-time performance with planning time included, confirming feasibility for autonomous operation.
• The method successfully captured spatial correlations and adapted to variable-resolution measurements, even with limited field of view.