Harpia: A hybrid system for agricultural UAV missions

Abstract

The present paper introduces Harpia, a hybrid artificial intelligent planning system for UAVs. Harpia aims to execute tasks for agricultural applications with minimum human intervention. The mission's execution is on-board, running under the Robotic Operating System and executing re-planning for tasks and path planning with obstacle avoidance. The re-planning can happen after mission changes in real-time or unpredictable UAV behavior. It combines Planning Domain Definition Language for task planning, Bayesian Network to evaluate mission execution, and K-Nearest Neighbors algorithm to select a path planner. Thus, Harpia's novelty focuses on robustness for autonomously planning and re-planning the sequence of tasks and trajectories to regions of interest. The main contributions include an autonomous system architecture for planning missions with minimum human intervention, no boundary by specific tasks, and computationally simple for operating within non-convex scenarios. The computational tests report results for 21 simulated scenarios, where Harpia handled all situations properly, e.g., making decisions about task re-planning with 97.57% accuracy based on battery health and choosing the better path-planning for each case with at least 95% of accuracy.