Exploration-Based Planning for Multiple-Target Search with Real-Drone Results.

Area of Science:

• Robotics
• Artificial Intelligence
• Computer Vision

Background:

• Drones are increasingly used for search and surveillance tasks.
• Efficiently locating unknown static targets is a significant challenge in drone operations.
• Existing methods like lawnmower and active-search patterns have limitations.

Purpose of the Study:

• To develop and evaluate a novel, efficient drone search strategy for localizing an unknown number of static targets.
• To combine exploration and target refinement in a receding-horizon planning framework.
• To demonstrate the effectiveness of the proposed method against baseline approaches.

Main Methods:

• Utilizing a multi-target particle filter to update target intensity based on imperfect measurements.
• Implementing a receding-horizon planner that optimizes drone movement for exploration and refinement.
• Integrating an obstacle-avoidance controller for waypoint navigation.
• Validating the approach through simulations and a real-world experiment with a Parrot Mambo drone.

Main Results:

• The proposed receding-horizon planner significantly improved search efficiency compared to lawnmower and active-search baselines.
• The system successfully localized targets using on-board image segmentation for measurements.
• The framework demonstrated robust performance in both simulated and real-world environments.
• Confidently localized targets were dynamically removed from the search space.

Conclusions:

• The novel receding-horizon planning framework offers a superior approach for drone-based static target search and localization.
• The integration of multi-target particle filtering and exploration-refinement objectives enhances search efficiency.
• The demonstrated success in real-world experiments highlights the practical applicability of the proposed method.