Paulo Drews-Jr

Federal University of Rio Grande | FURG · Center for Computer Science - C ³

Deep Reinforcement Learning (DRL) has emerged as a promising approach to enhancing motion control and decision-making through a wide range of robotic applications. While prior research has demonstrated the efficacy of DRL algorithms in facilitating autonomous mapless navigation for aerial and terrestrial mobile robots, these methods often grapple w...

Semantic Bird's Eye View (BEV) maps offer a rich representation with strong occlusion reasoning for various decision making tasks in autonomous driving. However, most BEV mapping approaches employ a fully supervised learning paradigm that relies on large amounts of human-annotated BEV ground truth data. In this work, we address this limitation by p...

Special Session: Hybrid and Convertible Unmanned Aerial Vehicles—Environments that have structures in multiple domains are a great challenge to performing inspection and monitoring tasks, often requiring several robots to perform in these ambient. Hybrid vehicles can become an alternative to the use of several devices. Especially, industrial marine...

Special Session: Sensing and Perception for Robotics Analyzing and monitoring underwater structures is still a challenge for humanity. A Remotely Operated Vehicle (ROV) equipped with sonar is a feasible alternative for assessing structures in harsh environments. However, the cost associated with this equipment is high. Therefore, this work presents...

Special Session: Hybrid and Convertible Unmanned Aerial Vehicles.-This paper describes the development of a sealing capsule for attaching the instrumentation required for a Hybrid Unmanned Aerial Underwater Vehicle (HUAUV) to perform tasks autonomously. The designed capsule aims to protect the electronic components in the underwater environment ens...

Quadrotors have been more frequently used in different areas, from aerial photography to drug delivery in medical emergencies. These vehicles have high maneuverability, which makes them suitable for carrying out missions that humans would not be able to do due to physical constraints. They can be used in inhospitable environments where the physical...

A expansão do histograma é um procedimento bastante utilizado no realce de imagens subaquáticas. Esta operação produz imagens com melhoria na cor e contraste. Entretanto, imagens com acentuado desequilíbrio entre os canais de cores geram regiões saturadas na expansão. Neste trabalho, propõe-se metodologia para evitar estas áreas saturadas. O método...

Methods for correcting errors generated by Optical Character Recognition (OCR) system are being developed for a long time, with interesting results in their applications. However, these methods tend to work only on data with words that are part of an existing language and with a large semantic relationship between each word in the text. In this wor...

This work studies the trajectory planning of an unmanned hybrid aerial-underwater vehicle (HUAUV) called Hydrone, which is being developed by the Intelligent Robotics and Automation Group (NAUTEC) of the Federal University of Rio Grande (FURG). This study presents a new trajectory planning algorithm, based on closed-loop rapidly exploring random tr...

O Aprendizado por Reforço (RL) tem se mostrado altamente eficaz em jogos eletrônicos e tarefas de controle contínuo. Entretanto, RL apresenta dificuldades em lidar com observações de alta dimensionalidade, como imagens brutas de pixels. É amplamente aceito que políticas de RL baseadas em estado físico, como medições de sensores a laser, geralmente...

This work presents a study on parallel and distributional deep reinforcement learning applied to the mapless navigation of UAVs. For this, we developed an approach based on the Soft Actor-Critic method, producing a distributed and distributional variant named PDSAC, and compared it with a second one based on the traditional SAC algorithm. In additi...

Deep Reinforcement Learning (Deep-RL) techniques for motion control have been continuously used to deal with decision-making problems for a wide variety of robots. Previous works showed that Deep-RL can be applied to perform mapless navigation, including the medium transition of Hybrid Unmanned Aerial Underwater Vehicles (HUAUVs). These are robots...

This paper presents a cross-domain and cross-view framework for underwater robot localisation, which does not require any Global Positioning System (GPS) information. The proposed localisation method uses colour aerial images and underwater acoustic images to estimate the robot’s position. The method identifies the correlation among images from dis...

This work deals with the closed-loop altitude control of a hybrid unmanned aerial-underwater vehicle (HUAUV) during media transition. Researches about HUAUVs are important for hard-to-reach operations. There are already several studies developed for unmanned vehicles, either in air or underwater. However, research on HUAUVs is a recent topic. There...

Uncertainty estimation is crucial in safety-critical settings such as automated driving as it provides valuable information for several downstream tasks including high-level decision-making and path planning. In this work, we propose EvCenterNet, a novel uncertainty-aware 2D object detection framework utilizing evidential learning to directly estim...

Bird's-Eye-View (BEV) semantic maps have become an essential component of automated driving pipelines due to the rich representation they provide for decision-making tasks. However, existing approaches for generating these maps still follow a fully supervised training paradigm and hence rely on large amounts of annotated BEV data. In this work, we...

Robots have been increasingly used in applications involving welding of large metal structures, such as the naval industry, ensuring higher efficiency and repeatability at lower costs. However, inadequate communication between the robotics and the welding system can lead to internal and surface defects in the final product. Problems that occur duri...

This paper introduces novel deep reinforcement learning (Deep-RL) techniques using parallel distributional actor-critic networks for navigating terrestrial mobile robots. Our approaches use laser range findings, relative distance, and angle to the target to guide the robot. We trained agents in the Gazebo simulator and deployed them in real scenari...

Underwater images suffer from degradation caused by water turbidity, light attenuation and color casting. An image enhancement procedure improves the perception and the analysis of the objects in the scene. Recent works based on deep learning approaches require synthetically paired datasets to train their models. In this work, it is present a self-...

Sonar sensors are an important source of data for understanding underwater environments because the sound is invariant to water turbidity, as well as illumination condition changes. This work reconstructs the surface of underwater structures and quantifies the spatial variations between multiple readings using data acquired by a 2D Mechanical Scann...

Este artigo apresenta um framework multi domínio e multi perspectiva para localização de robôs submarinos, que não requer nenhuma informação do Sistema de Posicionamento Global (GPS). O método de localização proposto usa imagens aéreas coloridas e imagens acústicas subaquáticas para estimar a posição do robô. O método identifica a correlação entre...

Previous works showed that Deep-RL can be applied to perform mapless navigation, including the medium transition of Hybrid Unmanned Aerial Underwater Vehicles (HUAUVs). This paper presents new approaches based on the state-of-the-art actor-critic algorithms to address the navigation and medium transition problems for a HUAUV. We show that a double...

Unmanned Aerial Vehicles (UAV) have been standing out due to the wide range of applications in which they can be used autonomously. However, they need intelligent systems capable of providing a greater understanding of what they perceive to perform several tasks. They become more challenging in complex environments since there is a need to perceive...

Deterministic and Stochastic techniques in Deep Reinforcement Learning (Deep-RL) have become a promising solution to improve motion control and the decision-making tasks for a wide variety of robots. Previous works showed that these Deep-RL algorithms can be applied to perform mapless navigation of mobile robots in general. However, they tend to us...

O Robot Operating System (ROS) é um dos mais populares softwares voltados ao desenvolvimento e pesquisa em robótica. Entretanto, estudos têm demonstrado diversos problemas de segurança em sua estrutura. Este trabalho avalia a aplicação de técnicas de detecção de intrusão no reconhecimento de ataques de injeção de dados nesses sistemas. Um modelo fo...

Images acquired during underwater activities suffer from environmental properties of the water, such as turbidity and light attenuation. These phenomena cause color distortion, blurring, and contrast reduction. In addition, irregular ambient light distribution causes color channel unbalance and regions with high-intensity pixels. Recent works relat...

Images acquired during underwater activities suffer from environmental properties of the water, such as turbidity and light attenuation. These phenomena cause color distortion, blurring, and contrast reduction. In addition, irregular ambient light distribution causes color channel unbalance and regions with high-intensity pixels. Recent works relat...

A busca pelo desenvolvimento de novas tecnologias impulsa grandes desafios. Exemplo disto refere-se ao desenvolvimento de tarefas correlatas aos robôs móveis híbridos. In order to study and overcome these challenges, the present work seeks to establish an approach based on Deep Reinforcement Learning (Deep-RL) para navegação autônoma de um tipo esp...

Reinforcement Learning (RL) has presented an impressive performance in video games through raw pixel imaging and continuous control tasks. However, RL performs poorly with high-dimensional observations such as raw pixel images. It is generally accepted that physical state-based RL policies such as laser sensor measurements give a more sample-effici...

Cross-view image matches have been widely explored on terrestrial image localization using aerial images from drones or satellites. This study expands the cross-view image match idea and proposes a cross-domain and cross-view localization framework. The method identifies the correlation between color aerial images and underwater acoustic images to...

In the last decade, a great effort has been employed in the study of Hybrid Unmanned Aerial Underwater Vehicles, robots that can easily fly and dive into the water with different levels of mechanical adaptation. However, most of this literature is concentrated on physical design, practical issues of construction, and, more recently, low-level contr...

Recently, Deep Convolutional Neural Networks have been successfully applied to various robotics problems, such as robot vision and simultaneous localization and mapping. Among these, siamese and triplet networks have obtained great traction in intra-domain matching. However, it is impossible to directly use these networks in cross-domain problems....

Cross-view image matches have been widely explored on terrestrial image localization using aerial images from drones or satellites. This study expands the cross-view image match idea and proposes a cross-domain and cross-view localization framework. The method identifies the correlation between color aerial images and underwater acoustic images to...

This paper presents a novel deep reinforcement learning-based system for 3D mapless navigation for Unmanned Aerial Vehicles (UAVs). Instead of using an image-based sensing approach, we propose a simple learning system that uses only a few sparse range data from a distance sensor to train a learning agent. We based our approaches on two state-of-art...

Modern imaging devices can capture faithful color and characteristics of natural and man-made scenes. However, there exist conditions in which the light radiated by objects cannot reach the camera’s lens or it is naturally degraded. Thus, the resulting captured images suffer from color loss. This article addresses the problem of underwater image re...

This paper presents a novel deep reinforcement learning-based system for 3D mapless navigation for Unmanned Aerial Vehicles (UAVs). Instead of using a image-based sensing approach, we propose a simple learning system that uses only a few sparse range data from a distance sensor to train a learning agent. We based our approaches on two state-of-art...

In the last decade, a great effort has been employed in the study of Hybrid Unmanned Aerial Underwater Vehicles, robots that can easily fly and dive into the water with different levels of mechanical adaptation. However, most of this literature is concentrated on physical design, practical issues of construction, and, more recently, low-level contr...

Image segmentation is an important step in many computer vision and image processing algorithms. It is often adopted in tasks such as object detection, classification, and tracking. The segmentation of underwater images is a challenging problem as the water and particles present in the water scatter and absorb the light rays. These effects make the...

Realizar a navegação entre meios é uma tarefa desafiadora para robôs móveis híbridos, especialmente em cenários com obstáculos. Este trabalho apresenta uma abordagem baseada em aprendizado por reforço profundo (Deep-RL) para navegação autônoma de um tipo específico de robô móvel híbrido: um Veículo Híbrido Tipo Ar-Água (HUAUV). A abordagem proposta...