| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 67.41 MB | Adobe PDF |
Autores
Resumo(s)
Osrobôsdenavegaçãoautónomasãocadavezmaisutilizadosnaagricultura, logística
e indústria de fabricação. A impressão 3D, uma técnica comum de fabrico aditivo, tem
ganho relevância nos últimos anos devido à sua flexibilidade e eficiência. A logística
de peças fabricadas pode ser suportada por Robôs Móveis Autónomos (AMRs), mel
horando o fluxo de materiais. No contexto da impressão 3D, é abordada a navegação
autónoma por objetivos em ambientes industriais, com foco em algoritmos de local
ização e mapeamento simultâneo (SLAM). Estes algoritmos constituem um elemento
fundamental dos robôs móveis autónomos, permitindo uma navegação fiável em am
bientes desconhecidos e sem acesso a GPS. Apesar da grande diversidade de soluções
SLAMexistentes,asavaliaçõescomparativasnaliteraturasão, emgeral, limitadasaum
reduzido número de algoritmos e conjuntos de dados, dificultando uma seleção infor
mada. Este trabalho propõe umametodologia deavaliação comparativa em duasfases
para colmatar esta limitação. Numa primeira fase, é introduzido um sistema de clas
sificação à escala alargada que agrega dados quantitativos de desempenho reportados
na literatura. Os algoritmos SLAM são então classificados com base na sua precisão
documentada e na abrangência das comparações, recorrendo a um modelo estatístico
ponderado. Arobustez daclassificação é reforçada através da variação sistemática dos
parâmetros e da análise da frequência de posicionamento no ranking. Numa segunda
fase, a eficácia da metodologia é validada através da avaliação direta de algoritmos se
lecionados num conjunto de dados comum, o NTU-VIRAL, utilizando o erro absoluto
de trajetória como métrica principal. Os algoritmos SLAM identificados são avaliados
qualitativamentequantoàsuaaplicabilidadenumambientereal. O3DMoBot,umrobô
móvel de locomoção diferencial para aplicações de impressão 3D, é configurado para
operar numambienteinterior, ondeétestada anavegaçãoautónomaentrelocalizações
predefinidas. A navegação autónoma depende de um algoritmo SLAM adequado, se
lecionado com base nosresultados das avaliações anteriormente desenvolvidas. Os re
sultados evidenciam uma forte consistência entre a classificação baseada na literatura
e a avaliação no conjunto de dados comum, e fornecem uma análise da aplicabilidade
dosalgoritmos melhorclassificados no3DMoBotintegradonumsistemadenavegação
baseado em ROS2.
Autonomousnavigationrobotsareincreasinglyusedinagriculture,logistics,andman ufacturing. 3D printing, a common additive manufacturing technique, has gained relevance in recent years due to its flexibility and efficiency. The logistics of manu factured parts can be supported by Autonomous Mobile Robots (AMRs), improving material flow. In the context of 3D printing, we addresses the autonomous objective based navigation in industrial environments with a focus on Simultaneously Localiza tion And Mapping (SLAM) algorithms. These algorithms are a core component of autonomous mobile robots, enabling reliable navigation in unknown and GPS-denied environments. Despite the large number of available SLAM solutions, comparative evaluations in the literature are typically limited to a small subset of algorithms and datasets, making informed selection difficult. This work proposes a two-stage com parative evaluation framework to address this limitation. First, a large-scale ranking methodology is introduced that aggregates quantitative performance data from the literature. SLAM algorithms are then scored and ranked based on documented accu racy and comparison coverage using a weighted statistical model. The robustness of this ranking is improved through systematic parameter variation and frequency-based ranking analysis. In the secondstage, the effectiveness of the proposedmethodologyis validated through direct evaluation of selected SLAM algorithms on acommonbench mark, the NTU-VIRAL dataset, using absolute trajectory error as the primary metric. Theidentified SLAMalgorithms are qualitatively evaluated for their applicability on a real-world environment. The 3DMoBot, a differential-drive mobile robot platform for 3D printing applications, is setup for operation within an indoor-environment where autonomous navigation between predefined locations is tested. The autonomous nav igation relies on a suitable SLAM algorithm which is selected based on the results of the previously elaborated evaluations. The results demonstrate strong consistency be tween the literature-based ranking and the benchmark-based evaluation, and provide an analysis of the applicability the of the highest ranking algorithms on the 3DMoBot integrated within a ROS2 based navigation system.
Autonomousnavigationrobotsareincreasinglyusedinagriculture,logistics,andman ufacturing. 3D printing, a common additive manufacturing technique, has gained relevance in recent years due to its flexibility and efficiency. The logistics of manu factured parts can be supported by Autonomous Mobile Robots (AMRs), improving material flow. In the context of 3D printing, we addresses the autonomous objective based navigation in industrial environments with a focus on Simultaneously Localiza tion And Mapping (SLAM) algorithms. These algorithms are a core component of autonomous mobile robots, enabling reliable navigation in unknown and GPS-denied environments. Despite the large number of available SLAM solutions, comparative evaluations in the literature are typically limited to a small subset of algorithms and datasets, making informed selection difficult. This work proposes a two-stage com parative evaluation framework to address this limitation. First, a large-scale ranking methodology is introduced that aggregates quantitative performance data from the literature. SLAM algorithms are then scored and ranked based on documented accu racy and comparison coverage using a weighted statistical model. The robustness of this ranking is improved through systematic parameter variation and frequency-based ranking analysis. In the secondstage, the effectiveness of the proposedmethodologyis validated through direct evaluation of selected SLAM algorithms on acommonbench mark, the NTU-VIRAL dataset, using absolute trajectory error as the primary metric. Theidentified SLAMalgorithms are qualitatively evaluated for their applicability on a real-world environment. The 3DMoBot, a differential-drive mobile robot platform for 3D printing applications, is setup for operation within an indoor-environment where autonomous navigation between predefined locations is tested. The autonomous nav igation relies on a suitable SLAM algorithm which is selected based on the results of the previously elaborated evaluations. The results demonstrate strong consistency be tween the literature-based ranking and the benchmark-based evaluation, and provide an analysis of the applicability the of the highest ranking algorithms on the 3DMoBot integrated within a ROS2 based navigation system.
Descrição
Palavras-chave
SLAM Robôs móveis autónomos Comparação de algoritmos Navegação Planeamento Impressão 3D
Contexto Educativo
Citação
Editora
Licença CC
Sem licença CC