An Information-Motivated Exploration Agent to Locate Stationary Persons with Wireless Transmitters in Unknown Environments.

Daniel Barry,Andreas Willig,Graeme Woodward

doi:10.3390/s21227695

Abstract

Unmanned Aerial Vehicles (UAVs) show promise in a variety of applications and recently were explored in the area of Search and Rescue (SAR) for finding victims. In this paper we consider the problem of finding multiple unknown stationary transmitters in a discrete simulated unknown environment, where the goal is to locate all transmitters in as short a time as possible. Existing solutions in the UAV search space typically search for a single target, assume a simple environment, assume target properties are known or have other unrealistic assumptions. We simulate large, complex environments with limited a priori information about the environment and transmitter properties. We propose a Bayesian search algorithm, Information Exploration Behaviour (IEB), that maximizes predicted information gain at each search step, incorporating information from multiple sensors whilst making minimal assumptions about the scenario. This search method is inspired by the information theory concept of empowerment. Our algorithm shows significant speed-up compared to baseline algorithms, being orders of magnitude faster than a random agent and 10 times faster than a lawnmower strategy, even in complex scenarios. The IEB agent is able to make use of received transmitter signals from unknown sources and incorporate both an exploration and search strategy.

Highlights

In the last few years, unmanned Aerial Vehicles (UAVs) spurred substantial interest, as they can improve the delivery of existing services or enable provision of new services in a wide range of fields, including logistics [1], search and rescue (SAR) [2,3,4] public safety communications [5,6], infrastructure monitoring [7], precision agriculture [8,9], forestry [10,11], and telecommunications [12,13]
We describe the design of a novel search algorithm, which is loosely based on the information-theoretic concept of empowerment, and which incorporates limited assumptions for properties of the wireless channel
In the approach described by Lanillos et al [30], we see a more similar algorithm to the one we describe in this paper, where they describe updating a Bayesian model and the UAV agent maximizes the estimated probability of locating a target given an action sequence

Summary

Introduction

Rodríguez-Gonzálvez and DiegoIn the last few years, unmanned Aerial Vehicles (UAVs) spurred substantial interest, as they can improve the delivery of existing services or enable provision of new services in a wide range of fields, including logistics [1], search and rescue (SAR) [2,3,4] public safety communications [5,6], infrastructure monitoring [7], precision agriculture [8,9], forestry [10,11], and telecommunications [12,13].In this paper, we explore the use of UAVs in SAR scenarios in an unknown and possibly large terrain, with the intention of reducing time for locating victims. We assume that the UAV is equipped with appropriate sensors to detect persons, e.g., based on a downward-facing camera using visible light or infrared These sensors allow the UAV to decide the presence or absence of a person only in a relatively small area determined by the visual angle of the camera and the flying height of the UAV. With such a camera alone, to maximize the certainty that all persons will be located, the UAV would have to pick a path that is “dense”, i.e., which guarantees that each point is observed at least once through the camera (e.g., a ’lawnmower’ path). We do not assume that the searching UAV has any a priori knowledge about the specific wireless technologies that any person may be using or channel properties, we only assume that the UAV is able to detect transmissions in a given frequency range, without

Objectives

Results

Discussion

Conclusion