Real-Time Sampling-based Online Planning for Drone Interception

TL;DR

The work tackles real-time, time-efficient interception of a moving target by a defense drone in cluttered, uncertain environments. It combines sampling-based trajectory generation with a neural network planning policy to replace expensive nonlinear optimization, enabling parallel exploration of multiple candidate target positions and rapid selection of the minimum-time feasible path. A multi-fidelity reinforcement learning framework assesses trajectory feasibility and refines time allocations, improving robustness to prediction noise and dynamics. Validated in simulation and real-world tests, the approach achieves high-rate online replanning (around 10 Hz) and demonstrates effective interception despite imperfect target predictions, suggesting strong practical impact for agile autonomous defense systems.

Abstract

This paper studies high-speed online planning in dynamic environments. The problem requires finding time-optimal trajectories that conform to system dynamics, meeting computational constraints for real-time adaptation, and accounting for uncertainty from environmental changes. To address these challenges, we propose a sampling-based online planning algorithm that leverages neural network inference to replace time-consuming nonlinear trajectory optimization, enabling rapid exploration of multiple trajectory options under uncertainty. The proposed method is applied to the drone interception problem, where a defense drone must intercept a target while avoiding collisions and handling imperfect target predictions. The algorithm efficiently generates trajectories toward multiple potential target drone positions in parallel. It then assesses trajectory reachability by comparing traversal times with the target drone's predicted arrival time, ultimately selecting the minimum-time reachable trajectory. Through extensive validation in both simulated and real-world environments, we demonstrate our method's capability for high-rate online planning and its adaptability to unpredictable movements in unstructured settings.

Figures (9)

• Figure 1: Overview of the proposed algorithm: (top left) simulation environment for drone interception experiments; (top right) generation of the point-mass path towards target drone position candidates; (bottom left) extraction of waypoints along these candidate paths; (bottom right) parallel optimization of the trajectories along the waypoints.
• Figure 2: (a) Neural network planning policy consisting of two gated recurrent units, attention, and variational autoencoder. The model outputs time allocations from the sequence of prescribed waypoints. (b) Relative total trajectory time of the planning policy output compared to the minimum snap method. The reinforcement learning (MFRL) model outputs faster time allocations, shifting the output distribution to the left.
• Figure 3: Utilizing a multi-resolution occupancy map for efficient A* search, where the lattice size increases with the distance between the target and defense drones.
• Figure 4: Iterative traversal time adaptation: (a) Time scaling method preserves trajectory shape while reducing speed. (b) Iterative application finds optimal scaling factor $\alpha_{\text{goal}}$ to match target time.
• Figure 5: Tracking error depending on the scale factor. Shading indicates standard deviation.