Occlusion-Aware Ground Target Search by a UAV in an Urban Environment

TL;DR

This work addresses searching for a moving POI on an urban road network with occlusions using a UAV modeled as a variable-speed Dubins vehicle. It introduces probabilistic visibility volumes represented as voxel-based matrices and couples them with a recursive Bayesian estimator to update target state distributions; planning leverages iterative deepening A* with a heuristic based on future reachable VV states and a max-pooling operation to balance horizon lengths. Key contributions include (i) a matrix-based probabilistic VV formalism, (ii) a forward-looking heuristic that uses reachability and VV convolution to bound future view probability, and (iii) a variable-timestep planner achieved via max-pooling to reduce search depth. Numerical studies show the proposed approach outperforms a static VV-based baseline and a lawnmower pattern in dense urban scenarios, particularly under higher false alarm rates, demonstrating improved search efficiency and robustness. The framework supports real-time planning and can be extended with GPU-accelerated matrix computations and integrated switching to tracking once the POI is localized.

Abstract

This paper considers the problem of searching for a point of interest (POI) moving along an urban road network with an uncrewed aerial vehicle (UAV). The UAV is modeled as a variable-speed Dubins vehicle with a line-of-sight sensor in an urban environment that may occlude the sensor's view of the POI. A search strategy is proposed that exploits a probabilistic visibility volume (VV) to plan its future motion with iterative deepening $A^\ast$. The probabilistic VV is a time-varying three-dimensional representation of the sensing constraints for a particular distribution of the POI's state. To find the path most likely to view the POI, the planner uses a heuristic to optimistically estimate the probability of viewing the POI over a time horizon. The probabilistic VV is max-pooled to create a variable-timestep planner that reduces the search space and balances long-term and short-term planning. The proposed path planning method is compared to prior work with a Monte-Carlo simulation and is shown to outperform the baseline methods in cluttered environments when the UAV's sensor has a higher false alarm probability.

Figures (18)

• Figure 1: Example of the environment where a UAV tracks a point of interest, left: 2D view and right: 3D view.
• Figure 2: A road network in an urban environment and the types of intersection in the environment.
• Figure 3: A probabilistic visibility volume for a "U" shaped road network around a single building.
• Figure 4: A matrix representing the probabilistic visibility for a "U" shaped road network.
• Figure 5: The reachability of the variable-speed Dubins vehicle for different initial headings $\psi_0$.