Online Informative Path Planning for Active Classification on UAVs

TL;DR

This paper introduces an informative path planning algorithm for UAVs to efficiently classify weeds in agriculture by optimizing trajectories based on information gain, combining global viewpoint selection with evolutionary optimization.

Contribution

It presents a novel IPP algorithm that integrates information-theoretic objectives with continuous trajectory optimization for UAV-based active classification.

Findings

Outperforms standard lawnmower coverage in simulation

Effectively balances information gain and flight constraints

Provides a flexible framework for UAV active classification

Abstract

We propose an informative path planning (IPP) algorithm for active classification using an unmanned aerial vehicle (UAV), focusing on weed detection in precision agriculture. We model the presence of weeds on farmland using an occupancy grid and generate plans according to information-theoretic objectives, enabling the UAV to gather data efficiently. We use a combination of global viewpoint selection and evolutionary optimization to refine the UAV's trajectory in continuous space while satisfying dynamic constraints. We validate our approach in simulation by comparing against standard "lawnmower" coverage, and study the effects of varying objectives and optimization strategies. We plan to evaluate our algorithm on a real platform in the immediate future.