Robot arácnido inteligente para detección de minas antipersonales metálicas en terreno irregular
Date
2020-12-07
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Pontificia Universidad Católica del Perú
Abstract
Debido a su bajo costo y a su simplicidad, enterrar minas antipersonales en el
campo de batalla ha sido, lastimosamente, una estrategia bélica muy común en el
siglo pasado. Sin embargo, la contaminación de tierras con minas antipersonales
es una problemática aún vigente en la actualidad, ya que más de sesenta países
siguen contaminados con estos explosivos. El desminado humanitario busca
erradicar todas las minas restantes en el mundo. Lamentablemente, el método
más utilizado actualmente para ejecutar labores de desminado es hacerlo de forma
manual, debido al elevado costo que representan los sistemas autónomos actuales.
Aunque existen propuestas de robots móviles que asisten estas labores, la mayoría
de ellas carece de versatilidad de desplazamiento sobre terreno irregular, ya que
utilizan ruedas para permitir el movimiento de estos sistemas.
En esta tesis, se propone el diseño de un robot arácnido para el traslado de un
sensor que permite la detección de minas antipersonales en terreno irregular.
En el ámbito mecánico, el robot tiene cuatro patas y presenta simetría sagital.
Los elementos electrónicos están alojados dentro del tórax, que se encuentra en el
centro del robot. El detector de minas se sujeta desde el tórax apuntando hacia
abajo. Se hizo el análisis dinámico del mecanismo las patas, con su respectiva
simulación cinemática y cinética. Además, se realizaron dos simulaciones de
esfuerzos en la estructura del robot mediante el método de elementos finitos.
Con respecto al diseño electrónico, se desarrollaron cálculos para los circuitos de
las fuentes de voltaje internas del robot. Se verificó la estabilidad de dichas fuentes
frente a ruidos eléctricos, ejecutando una simulación de los circuitos reguladores
de voltaje agregando ruido sinusoidal. Se realizó también la selección de los demás
componentes electrónicos del robot.
El ámbito de control es el aspecto más desarrollado de la presente tesis. Se plantea
un algoritmo innovador, que fusiona perspectivas de aprendizaje de máquina e
inteligencia artificial bioinspirada, para permitir que el robot aprenda a caminar.
De esta manera, se espera que el robot mejore su desempeño con el tiempo, a
medida que vaya obteniendo más experiencia y haya recolectado más datos de su
entorno. El algoritmo propuesto permite entrenar un agente inteligente para
maximizar una métrica de recompensa, pero sin aplicar técnicas de aprendizaje
por refuerzo, que son usualmente utilizadas en este tipo de problemas.
Due to its low cost and simplicity, burying anti-personnel landmines on the battlefield has unfortunately been a ubiquitous war strategy in the last century. However, land contamination with anti-personnel landmines is yet in force today since more than sixty countries are still polluted with these explosives. Humanitarian demining seeks to eradicate all remaining landmines worldwide. Unfortunately, today's most widely used demining method is manual due to the high cost of existing autonomous systems. Although there are proposals for mobile robots that assist these tasks, most of them lack the versatility of movement on uneven terrain since they use wheels for their locomotion. In this thesis, an arachnid robot's design is proposed to carry a sensor that allows the detection of anti-personnel landmines on uneven terrain. In the mechanical domain, the robot has four legs and has sagittal symmetry. The electronic components are housed within the thorax, which is located in the center of the robot. The mine detector is held by the thorax, pointing downwards. A dynamic analysis of the leg mechanism was made, with its respective kinematic and kinetic simulation. Also, two stress simulations were made for the robot structure using the finite element method (FEM). Regarding the electronic design, calculations were performed for the circuits of the robot's internal voltage sources. The stability of these sources against electrical noise was verified. A simulation of the voltage regulator circuits' was developed to check their response while adding sinusoidal noise. Additionally, the selection of the other electronic components of the robot was made. The control domain is the most developed aspect of this thesis. An innovative algorithm is proposed, which merges machine learning and bio-inspired artificial intelligence perspectives to allow the robot to learn to walk. In this way, the robot improves its performance over time as it gains more experience and collects more data from its environment. The proposed algorithm allows training an intelligent agent to maximize a reward metric without applying reinforcement learning techniques, which are usually the way to solve this type of problem.
Due to its low cost and simplicity, burying anti-personnel landmines on the battlefield has unfortunately been a ubiquitous war strategy in the last century. However, land contamination with anti-personnel landmines is yet in force today since more than sixty countries are still polluted with these explosives. Humanitarian demining seeks to eradicate all remaining landmines worldwide. Unfortunately, today's most widely used demining method is manual due to the high cost of existing autonomous systems. Although there are proposals for mobile robots that assist these tasks, most of them lack the versatility of movement on uneven terrain since they use wheels for their locomotion. In this thesis, an arachnid robot's design is proposed to carry a sensor that allows the detection of anti-personnel landmines on uneven terrain. In the mechanical domain, the robot has four legs and has sagittal symmetry. The electronic components are housed within the thorax, which is located in the center of the robot. The mine detector is held by the thorax, pointing downwards. A dynamic analysis of the leg mechanism was made, with its respective kinematic and kinetic simulation. Also, two stress simulations were made for the robot structure using the finite element method (FEM). Regarding the electronic design, calculations were performed for the circuits of the robot's internal voltage sources. The stability of these sources against electrical noise was verified. A simulation of the voltage regulator circuits' was developed to check their response while adding sinusoidal noise. Additionally, the selection of the other electronic components of the robot was made. The control domain is the most developed aspect of this thesis. An innovative algorithm is proposed, which merges machine learning and bio-inspired artificial intelligence perspectives to allow the robot to learn to walk. In this way, the robot improves its performance over time as it gains more experience and collects more data from its environment. The proposed algorithm allows training an intelligent agent to maximize a reward metric without applying reinforcement learning techniques, which are usually the way to solve this type of problem.
Description
Keywords
Robots--Diseño y construcción, Detectores, Sensores inteligentes
Citation
Collections
Endorsement
Review
Supplemented By
Referenced By
Creative Commons license
Except where otherwised noted, this item's license is described as info:eu-repo/semantics/openAccess