In the current scenario of fast technological advancement, increasingly characterized by widespread adoption of Artificial Intelligence (AI)-driven tools, the significance of autonomous systems like chatbots has been highlighted. Such systems, which are proficient in addressing queries based on PDF files, hold the potential to revolutionize customer support and post-sales services in the automotive sector, resulting in time and resource optimization. Within this scenario, this work explores the adoption of Large Language Models (LLMs) to create AI-assisted tools for the automotive sector, assuming three distinct methods for comparative analysis. For them, broad assessment criteria are considered in order to encompass response accuracy, cost, and user experience. The achieved results demonstrate that the choice of the most adequate method in this context hinges on the selected criteria, with different practical implications. Therefore, this work provides insights into the effectiveness and applicability of chatbots in the automotive industry, particularly when interfacing with automotive manuals, facilitating the implementation of productive generative AI strategies that meet the demands of the sector.

The air pollution caused by greenhouse gas emissions, particularly carbon dioxide (CO${}_2$), is a significant environmental concern that impacts air quality and contributes to global warming. The transportation sector plays a pivotal role in this issue, being a major contributor to CO${}_2$ emissions. In light of this situation, this article proposes a methodology that utilizes a supervised learning algorithm to estimate CO${}_2$ emissions and compare vehicles fueled with ethanol and gasoline. Additionally, the solution adopts an online, unsupervised machine learning algorithm to identify data outliers and improve the confidence in the results. Furthermore, this work incorporates the concept of digital twins, using virtual models of vehicles to carry out more extensive pollution simulations and allowing the simulation of various types of vehicles and the modeling of realistic traffic scenarios. A supervised machine learning approach was adopted to infer emission data in the model, allowing more comprehensive and meaningful comparisons between real-world and simulated measurements. The performed analyses of pollution emissions for different speeds and sections of routes demonstrate that CO${}_2$ emissions from ethanol were significantly lower than those from gasoline, favoring more sustainable fuels even in combustion engine vehicles. Adopting cleaner fuels is perceived as crucial to mitigate the negative effects of climate change, with plant-based fuels like ethanol being crucial during the transition from fossil fuels to a more sustainable vehicular landscape.

The Internet of Things (IoT) paradigm encompasses computing and networking capabilities within electronic objects, acting as a fundamental development framework with vast potential for improving lives, enhancing industrial processes, and enabling real-time decision-making. However, as the number of connected objects increases, the infrastructure for processing and handling large volumes of data is highly impacted. In response, edge computing has been exploited as a way to bring the processing burden closer to the data sources, shifting the conventional data processing flow. As a result, a series of innovative machine learning applications has been developed for resource-constrained devices, such as microcontrollers, enabling efficient data processing on the edge and inaugurating the era of Tiny Machine Learning (TinyML). Nevertheless, although the benefits have proven promising in different scenarios, particularly when the flexibility of embedded models meets the requisites of unsupervised learning approaches, TinyML-based applications may need to perform real-time identification of data outliers, since they could waste resources and impair the expected model accuracy. In this context, this article proposes an innovative TinyML outlier detection and correction algorithm based on incremental learning. This algorithm was implemented in an OBD-II scanner as a proof of concept, where a microcontroller acquires real-time vehicle data to identify data outliers and perform necessary corrections, benefiting practical applications in multiple scenarios.

Among the main causes of road accidents, one of the most significant is related to driver distraction while driving, which is responsible for 18% of car accidents worldwide. This situation has demanded the development of mechanisms to automatically detect this dangerous behavior while driving. One of the computational solutions that has been considered viable to detect situations like this is the use of Convolutional Neural Networks (CNN), but some complex issues arise when deploying CNN models in microcontroller-based embedded devices with constrained processing and memory capabilities. In this context, this paper proposes a driver distraction detection system that achieves high accuracy (99.3%) and low latency (72ms) while requiring minimal computational resources (Peak- RAM of 164 KB and Flash of 52.7 KB). This solution exploiting Tiny Machine Learning (TinyML) algorithms was developed with the support of the Edge Impulse platform, used to perform the entire ML pipeline, from data pre-processing and ML model creation to deployment into an Arduino Portenta H7 board. By designing a driver assistance system that can be integrated into vehicles, it is expected that an affordable solution based on embedded machine learning is provided, tackling a real-world problem by potentially reducing accidents caused by driver distractions.

The Internet of Things (IoT) has significantly impacted various industries, particularly the automotive sector, due to the growing integration of IoT technologies in vehicles. As a result, the volume of data generated by vehicular sensors has increased, leading to a surge in studies focusing on driver behavior to enhance road safety and optimize transportation networks. However, traditional approaches to analyzing driver behavior have relied on supervised offline learning models, which are unsuitable for handling data streams in online learning environments. This study introduces an unsupervised online k-fix AutoCloud algorithm for detecting driver behavior patterns, leveraging the concepts of typicity and eccentricity while considering the historical-temporal relationships between samples. Furthermore, the algorithm autonomously and adaptively evolves without requiring a supervised training phase, making it compatible with the TinyML concept, encompassing Artificial Intelligence algorithms designed for low-power IoT devices. To validate the proposed method, a real case study was conducted over four days using a vehicle to compare the quantity and quality of clusters generated by the algorithm. The findings demonstrate the potential of the proposed approach for optimizing data processing with minimal computational power.