Antonio Navarrete Guzmán

El objetivo de esta investigación es desarrollar una plataforma que controle la cantidad de insulina necesaria para mantener los niveles de glucosa en pacientes con diabetes mellitus tipo 1. El estudio se basa en la implementación del modelo matemático de Sorensen y un controlador basado en aprendizaje por refuerzo llamado (Reinforcement Learning-Based Controller, RLBC) Deep Q-Learning en una tarjeta de desarrollo Raspberry Pi 2B, con una ejecución en tiempo real. Los resultados obtenidos son comparados cuantitativamente con un controlador robusto basado en µ-síntesis, concluyendo que el controlador basado en aprendizaje por refuerzo presenta un mejor desempeño. La implicación principal de esta investigación es la posibilidad de mejorar el control de la glucosa en pacientes con diabetes mellitus tipo 1 a través de un sistema automatizado y de bajo costo. La contribución original de este estudio radica en la integración de un modelo matemático complejo con hardware accesible para el control de la diabetes. En conclusión, este trabajo sienta las bases para el desarrollo de dispositivos de control de insulina más eficientes y accesibles, aunque se requiere investigación adicional para validar su efectividad en entornos clínicos.

DrChatPatin is an LLM-based assistant for differential diagnosis of common and rare diseases. It follows an explicit dialogue policy to elicit missing clinical details and returns a ranked top-5 list with confidence scores and rationales. A RAG mode retrieves evidence from 1,250 NORD rare-disease reports and surfaces them as 'Diseases cited'. The system runs on the Internet Computer Protocol blockchain with end-to-end AES-256 encryption. Evaluated over 56 NEJM benchmark cases (280 queries per branch), the RAG-enabled mode achieves 65.4% hit@5 ICD-10 category accuracy versus 60.0% without RAG and 31.7% for the GPT-4 reference. The RAG benefit is larger for rare diseases (+7.1 pp) than for common diseases (+2.6 pp).

Diabetes mellitus, a chronic condition affecting millions worldwide, necessitates continuous monitoring of blood glucose level (BGL). The increasing prevalence of diabetes has driven the development of non-invasive methods, such as electronic noses (e-noses), for analyzing exhaled breath and detecting biomarkers in volatile organic compounds (VOCs). Effective machine learning models require extensive patient data to ensure accurate BGL predictions, but previous studies have been limited by small sample sizes. This study addresses this limitation by employing conditional generative adversarial networks (CTGAN) to generate synthetic data from real-world tests involving 29 healthy and 29 diabetic participants, resulting in over 14,000 new synthetic samples. These data were used to validate machine learning models for diabetes detection and BGL prediction, integrated into a Tiny Machine Learning (TinyML) e-nose system for real-time analysis. The proposed models achieved an 86% accuracy in BGL identification using LightGBM (Light Gradient Boosting Machine) and a 94.14% accuracy in diabetes detection using Random Forest. These results demonstrate the efficacy of enhancing machine learning models with both real and synthetic data, particularly in non-invasive systems integrating e-noses with TinyML. This study signifies a major advancement in non-invasive diabetes monitoring, underscoring the transformative potential of TinyML-powered e-nose systems in healthcare applications.

This article explores the application of the Espressif ESP32 System-On-Chip (SoC) for managing vehicle dynamics through real-time digital proportional–integral (PI-like) control. We present the development of advanced driving assistance algorithms for Active Front Steering (AFS) and Rear Torque Vectoring (RTV) on this cost-effective, commercially available embedded system. Using digital PI-like control algorithms designed for AFS and RTV, the primary ESP32 board receives and processes steering signals, executing a discrete-time control model of the vehicle dynamic to enable dynamic adjustments to steering and torque. To enhance simulation realism, a secondary ESP32 is employed to generate the steering signal, effectively mimicking a steer-by-wire system via its analog output ports. This configuration facilitates the simulation and evaluation of control algorithms in a realistic test environment, ensuring enhanced vehicle dynamic stability and maneuverability under various conditions. Additionally, simulations are conducted using MATLAB 2023a and CarSim 2017.1 to compare the efficacy and benefits of the implementation. Our objective is to establish a platform for evaluating discrete controllers capable of real-time vehicle operation. This methodology accelerates and reduces the cost of improving vehicle system stability and responsiveness, enabling the immediate verification and fine-tuning of control parameters as needed.

This study addresses the control problem in Electric Vehicles (EVs), specifically focusing on designing a robust dynamic control system capable of estimating parametric uncertainties and external disturbances. The EV under consideration is equipped with an Active Front Steering (AFS) system, which introduces an incremental steering angle to the front wheels. Additionally, the vehicle significantly enhances the powertrain, employing Permanent Magnet Synchronous Motors (PMSMs) mounted on the left and right wheel axle shafts. This configuration not only provides longitudinal traction but also facilitates appropriate yaw torque control, similar to the functionality of a Rear Torque Vectoring (RTV) system. This paper demonstrates how integrating the AFS system and RTV within the framework of integral control ensures that the vehicle can accurately track a predetermined trajectory, even in the presence of environmental disturbances, parametric uncertainties, or unmodeled dynamics. However, the proposed control strategy relies on precise information from the EV, particularly lateral velocity, which is challenging to measure directly in practical applications. To address this issue, a robust nonlinear observer is proposed to reconstruct the lateral velocity. Furthermore, High-Order Sliding Mode (HOSM) estimators are employed to approximate the parametric uncertainties and external perturbations in finite-time. The controller’s performance is evaluated through simulations of a demanding double-step steer maneuver, conducted using Simulink ^® software.

The antilock braking system (ABS) is a mechatronic system that helps a driver maintain the maneuverability of a vehicle while braking by preventing wheel lock-ups. However, the design of high-performance controllers for this type of system is complicated because of its highly nonlinear dynamics. The problem becomes even more difficult to resolve when uncertainties in the parameters appear in its dynamics. In this paper, an ABS laboratory setup mimicking a quarter car model is considered. A modified high-order sliding mode (HOSM) controller using a proportional-integral-differential (PID) control as a sliding surface was designed. This controller provides a reference value of a tire slip. The proposed controller uses a tracking error to define the slip surface through the PID controller structure, and the modified HOSM controller holds the system on the previously designed slip surface. The closed-loop system stability has been proven in the sense of Lyapunov. Finally, the ABS laboratory setup allows for experimentally checking the performance of the modified HOSM controller using a PID-sliding surface, showing a considerable increase in the efficiency of the control system compared with a PID-like controller.

The antilock braking system (ABS) is an electromechanical device whose controller is challenging to design because of its nonlinear dynamics and parameter uncertainties. In this paper, an adaptive controller is considered under the assumption that the friction coefficient is unknown. A modified high-order sliding-mode controller is designed to enhance the controller performance. The controller ensures tracking of the desired reference and identifies the unknown parameter, despite parametric variations acting on the real system. The stability proof is done using the Lyapunov approach. Some numerical and experimental tests evaluate the controller on a mechatronic system that represents a quarter-car model.

This study considers the design of a modified high-order sliding mode (HOSM) controller using a PI sliding surface to the attitude control of a ground vehicle. A robust-modified HOSM controller is derived, so that the lateral velocity and yaw rate tracks the desired trajectory despite the environment actions acting on the ground vehicle and parameter variations. The stability is guaranteed with Lyapunovs stability theorem function. The performance of the dynamic controllers is evaluated using the CarSim simulator considering a challenging double steer maneuver.