Valor estadístico de la vida: Regresión lineal versus regresión simbólica.
Loading...
Date
2025
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad de Concepción
Abstract
El desarrollo de capacidades autónomas en RPAs requiere de entornos de validación seguros y reproducibles que permitan experimentar con algoritmos de decisión antes de su aplicación en vuelo real. En el ámbito académico, las limitaciones operativas y la ausencia de plataformas experimentales integradas dificultan este proceso. Frente a este problema, la presente memoria busca diseñar e implementar un ecosistema modular para la validación de algoritmos de decisión, basado en un autopiloto Pixhawk con firmware PX4 y una Raspberry Pi como computadora embebida.
La metodología se desarrolló en tres etapas i) implementación de un entorno de simulación SITL utilizando PX4 y Gazebo en una máquina virtual, ii) integración de la Raspberry Pi al ecosistema mediante MAVSDK-Python para ejecutar detección de estímulos y maniobras en modo de vuelo autónomo OFFBOARD y iii) transferencia de la arquitectura a un RPA físico mediante pruebas en un banco de ensayos. Este enfoque permitió contrastar el desempeño del sistema tanto en simulación como en hardware real.
Los resultados mostraron que SITL reproduce con buena precisión una misión de vuelo real, obteniendo un RMSE de 0.490 [m] respecto a los waypoints de referencia, frente a 0.623 [m] del vuelo físico. La Raspberry Pi logró modificar misión y ejecutar maniobras autónomas en ambos entornos. En el banco de ensayos, la estimación de la duración efectiva para las maniobras en modo OFFBOARD promedio 1.68 [s] para Roll, 1.73 [s] para Pitch y 2.96 [s] para Yaw, valores ligeramente menores a los definidos en el código (2 y 4 segundos). La ejecución de este mismo código en SITL generó discrepancias importantes, aunque el simulador generó maniobras más suaves, no respetó la lógica temporal programada, por lo que se evidencia que los algoritmos no son completamente transferibles entre entornos y deben tener ajustes específicos.
El ecosistema desarrollado permite validar de forma modular algoritmos complejos aplicados a RPAs, demostrando su viabilidad técnica y proporcionando una base para ensayos de mayor complejidad.
The development of autonomous capabilities in RPAs requires safe and reproducible validation environments that enable experimentation with decision-making algorithms prior to their application in real flight. In academic environments, operational constraints and the lack of integrated experimental platforms hinder this process. To address this problem, the present work aims to design and implement a modular ecosystem for validating autonomous decision-making algorithms, based on a Pixhawk autopilot running PX4 firmware and a Raspberry Pi as an embedded companion computer. The methodology was structured into three stages: i) implementation of a SITL simulation environment using PX4 and Gazebo in a virtual machine, ii) integration of the Raspberry Pi into the ecosystem through MAVSDK-Python to perform stimulus detection and autonomous maneuvers in OFFBOARD mode, and iii) transfer of the architecture to a physical RPA through bench-test experiments. This approach enabled a direct comparison of system performance in both simulation and real hardware. The results showed that SITL reproduces a real flight mission with good accuracy, achieving an RMSE of 0.490 [m] with respect to the reference waypoints, compared to 0.623 [m] obtained in the actual flight. The Raspberry Pi successfully modified missions and executed autonomous maneuvers in both environments. In the bench tests, the effective duration of OFFBOARD maneuvers averaged 1.68 [s] for roll, 1.73 [s] for pitch, and 2.96 [s] for yaw, values slightly lower than those defined in the code (2 and 4 seconds). Running this same code in SITL revealed notable discrepancies: although the simulator produced smoother maneuvers, it did not respect the programmed timing logic, indicating that algorithms are not fully transferable between environments and require specific adjustments. The ecosystem developed in this work enables modular validation of complex algorithms applied to RPAs, demonstrating its technical feasibility and providing a foundation for experiments of greater complexity.
The development of autonomous capabilities in RPAs requires safe and reproducible validation environments that enable experimentation with decision-making algorithms prior to their application in real flight. In academic environments, operational constraints and the lack of integrated experimental platforms hinder this process. To address this problem, the present work aims to design and implement a modular ecosystem for validating autonomous decision-making algorithms, based on a Pixhawk autopilot running PX4 firmware and a Raspberry Pi as an embedded companion computer. The methodology was structured into three stages: i) implementation of a SITL simulation environment using PX4 and Gazebo in a virtual machine, ii) integration of the Raspberry Pi into the ecosystem through MAVSDK-Python to perform stimulus detection and autonomous maneuvers in OFFBOARD mode, and iii) transfer of the architecture to a physical RPA through bench-test experiments. This approach enabled a direct comparison of system performance in both simulation and real hardware. The results showed that SITL reproduces a real flight mission with good accuracy, achieving an RMSE of 0.490 [m] with respect to the reference waypoints, compared to 0.623 [m] obtained in the actual flight. The Raspberry Pi successfully modified missions and executed autonomous maneuvers in both environments. In the bench tests, the effective duration of OFFBOARD maneuvers averaged 1.68 [s] for roll, 1.73 [s] for pitch, and 2.96 [s] for yaw, values slightly lower than those defined in the code (2 and 4 seconds). Running this same code in SITL revealed notable discrepancies: although the simulator produced smoother maneuvers, it did not respect the programmed timing logic, indicating that algorithms are not fully transferable between environments and require specific adjustments. The ecosystem developed in this work enables modular validation of complex algorithms applied to RPAs, demonstrating its technical feasibility and providing a foundation for experiments of greater complexity.
Description
Tesis presentada para optar al título de Ingeniero/a Civil Aeroespacial.
Keywords
Algoritmos computacionales, Simulación por computadores, Vuelos de entrenamiento