| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 19.25 MB | Adobe PDF |
Authors
Abstract(s)
A capacidade de se localizar com precisão é fundamental para robôs autônomos. Vários métodos voltados para a solução desse problema foram desenvolvidos ao longo do tempo, incluindo métodos clássicos, marcadores fiduciais e, mais recentemente técnicas de aprendizado de máquina (do inglês, machine learning, ML). Esse trabalho propõe diferentes técnicas de ML para abordar o problema da localização de robôs na competição RobotAtFactory 4.0. Esse estudo abrange desde testes das abordagens em sistemas embebidos, com foco na viabilidade da aplicação desses métodos, na exploração de diversos modelos e
abordagens usando ML, até a aplicação de modelos treinados em simulação em ambientes reais. Os resultados experimentais mostraram que os modelos podem ser executados em sistemas embebidos, e diversas técnicas obtiveram resultados com precisão milimétrica. Além disso, a aplicação direta de modelos treinados em simulação para ambiente real se apresentou promisora. Uma das principais vantagens dos modelos de ML é o desvinculo com a dependência do conhecimento prévio da posição exata dos marcadores fiducias.
The ability to accurately localize is a fundamental key for autonomous robots. Various methods to solve this problem have been developed, including classical methods, fiducial markers, and, more recently, machine learning (ML) techniques. This work proposes different ML techniques to address the issue of robot localization in the RobotAtFactory 4.0 competition. This study includes testing the approaches on embedded systems, focusing on the feasibility of applying these methods, exploring various ML models and approaches, and applying simulation-trained models in real environments. The experimental results demonstrated that the models could be executed on embedded systems, and several techniques achieved millimeter-level accuracy. Furthermore, the direct application of simulation-trained models to real environments showed promise. One of the main advantages of ML models is their independence from the need for prior knowledge of the exact positions of fiducial markers.
The ability to accurately localize is a fundamental key for autonomous robots. Various methods to solve this problem have been developed, including classical methods, fiducial markers, and, more recently, machine learning (ML) techniques. This work proposes different ML techniques to address the issue of robot localization in the RobotAtFactory 4.0 competition. This study includes testing the approaches on embedded systems, focusing on the feasibility of applying these methods, exploring various ML models and approaches, and applying simulation-trained models in real environments. The experimental results demonstrated that the models could be executed on embedded systems, and several techniques achieved millimeter-level accuracy. Furthermore, the direct application of simulation-trained models to real environments showed promise. One of the main advantages of ML models is their independence from the need for prior knowledge of the exact positions of fiducial markers.
Description
Mestrado de dupla diplomação com a UTFPR - Universidade Tecnológica Federal do Paraná
Keywords
Localização de robôs Machine-learning Marcadores fiducias Competição de robótica