Angel Ayala

Research interest in autonomous control of unmanned aerial vehicles (UAVs) has increased rapidly over the past decade. They are now widely used in civilian, military, and private areas. Applications include surveillance, search and rescue, and delivery tasks. More broadly, they excel at solving problems where a significant amount of space must be covered and traveled. However, using UAVs to solve navigation problems with full autonomy necessitates the mastering of complex subtasks. A solution that includes control, planning, localization, and mapping remains an open challenge. Object-goal navigation contains the same navigation problems where the main objective is to reach a target object. The search and identification of this target are central to the vehicle’s navigation. This requires an understanding of what it is and where it can be located to move around the scene. This paper presents a systematic literature review on object-goal navigation and its subtasks, using autonomous UAVs. Survey taxonomies were found for the tasks and methods behind navigation and target localization problems using UAVs. The review analyzed 67 articles found between 2011 and 2022. They were found in the ACM, IEEE Xplore, WebOfScience, Science Direct, and Scopus databases. This review revealed essential issues related to autonomous navigation task dependencies. Moreover, it highlighted gaps in UAV development and framework standardization. Open challenges for autonomous UAV control for object-goal navigation must address the research on finding methods for problems. For example, autonomy level and comparison metrics, considering safety, ethics, and legal implications.

One problem with cloud computing is that it may fail to meet the desired time limits for real-time applications. In this regard, fog computing paradigm has gained ground as it complements the cloud by providing nodes with processing and storage capabilities closer to the data generation level. However, this level of the architecture has limited resources, making it necessary to efficiently distribute the workload involved in applications, especially when employing deep learning models. One technique to achieve this is task offloading, which involves distributing inference tasks throughout the architecture. Nevertheless, it is also important to know the time required for these tasks to be carried out within the network in order to obtain the desired response. In this work, we propose a queue-based convolutional neural network that allows estimating the response time for a deep learning inference task. Preliminary results demonstrate a good fit to the behavior of the datasets used in the experiment.

In the era of big data, resource utilization is a critical consideration. This is particularly true in machine learning and computer vision, where the complexity of images often necessitates large volumes of data to train models effectively. However, data scarcity presents a significant obstacle in certain areas, such as human action recognition (HAR), particularly in real-world applications like violence detection in videos. This research responds to this challenge by recycling a preexisting human action recognition model to perform a new task - violence detection. The HAR model is used as a feature extractor and combined with classifiers to abstract information from the model and categorize videos into regular activities or violence. This approach allows for adequate performance with less data. Two prominent HAR models, Spatial-Temporal Graph Convolutional Network and Mobile Video Network, pretrained on the Kinetics400 and Kinetics600 datasets, respectively, are repurposed and paired with classifiers for this task. In addition, these models are evaluated on two violence datasets, UCF-Crime and Brazilian Violence, to assess their performance in differentiating between violent and non-violent behavior. Both models achieve high Area Under the Receiver Operating Characteristic scores, indicating their ability to discern between the positive and negative classes.

Autonomous control for unmanned aerial vehicles (UAVs) has rapidly raised research interest in the last decade. Widely used in civilian, military, and private areas, UAVs have been implemented in surveillance, search and rescue, and delivery tasks, solving problems where a significant space must be covered and traveled. However, using UAVs for navigation problems with full autonomy in control, planning, localization, and mapping tasks remains an open challenge. This paper presents a systematic literature review on object-goal navigation using autonomous UAVs. Object-goal navigation inherits the same navigation challenges, such as control and navigation, in addition to estimating the target's location. The object-search problem requires to understand what object must be found and where it can be located.In this regard, survey taxonomies were found for the tasks and methods behind navigation and target localization problems using UAVs. This review reveals essential issues related to autonomous navigation task dependencies and gaps in UAV development and framework standardization. Open challenges for autonomous UAV control under object-goal navigation must address the research on finding methods for problems, such as autonomy level and comparison metrics, considering safety, ethics, and legal implications.

Currently, artificial intelligence is in an important period of growth. Due to the technology boom, it is now possible to solve problems that could not be resolved previously. For example, through goal-driven learning, it is possible that intelligent machines or agents may be able to perform tasks without human intervention. However, this also leads to the problem of understanding the agent’s decision making. Therefore, explainable goal-driven learning attempts to eliminate this gap. This work focuses on the adaptability of two explainability methods in continuous environments. The methods based on learning and introspection proposed a probability value for success to explain the agent’s behavior. These had already been tested in discrete environments. The continuous environment used in this study is the car-racing problem. This is a simulated car racing game that forms part of the Python Open AI Gym Library. The agents in this environment were trained with the Deep Q-Network algorithm, and in parallel the explainability methods were implemented. This research included a proposal for carrying out the adaptation and implementation of these methods in continuous states. The adaptation of the learning method produced major changes, implemented through an artificial neural network. The obtained probabilities of both methods were consistent throughout the experiments. The probability result was greater in the learning method. In terms of computational resources, the introspection method was slightly better than its counterpart.

Early alert fire and smoke detection systems are crucial for daily and security management decision-making. Recent literature approaches are based on Deep Learning (DL) models. Efficient models are required for hardware-constrained systems, such as mobile devices, embedded systems, and robotics achieving high performance at low power consumption. For this research, we designed a novel specific-purpose model for fire and smoke recognition using still images and the study of state-of-the-art convolution techniques to improve the trade-off between accuracy and complexity. In this regard, the literature suggests that the inverted residual block, the depthwise and octave convolution techniques, reduces the model’s size and computation requirements working well by themselves. In this work, we propose the KutralNext architecture, an efficient model for single- and multi-label fire and smoke recognition tasks. Additionally, a more efficient architecture KutralNext+, demonstrates that those convolution techniques achieve a better trade-off combined, reaching an 84.36% average test accuracy in FireNet, FiSmo, and FiSmoA fire datasets. The KutralSmoke and FiSmo fire and smoke datasets attained an 81.53% average test accuracy. Furthermore, a previous fire and smoke recognition model considered, FireDetection, KutralNext uses 59% fewer parameters, and KutralNext+ requires 97% fewer flops and is 4x faster.

Diabetes is a disease that occurs when the body presents an uncontrolled level of glucose that is capable of damaging the retina, leading to permanent damage of the eyes or vision loss. When diabetes affects the eyes, it is known as diabetic retinopathy, which became a global medical problem among elderly people. The fundus oculi technique involves observing the eyeball to diagnose or check the pathology evolution. In this work, we implement a convolutional neural network model to process a fundus oculi image to recognize the eyeball structure and determine the presence of diabetic retinopathy. The model’s parameters are optimized using the transfer-learning methodology for mapping an image with the corresponding label. The model training and testing are performed with a dataset of medical fundus oculi images and a pathology severity scale present in the eyeball as labels. The severity scale separates the images into five classes, from a healthy eyeball to a proliferative diabetic retinopathy presence. The latter is probably a blind patient. Our proposal presented an accuracy of 97.78%, allowing for the confident prediction of diabetic retinopathy in fundus oculi images.

Currently, unmanned aerial vehicles, such as drones, are becoming a part of our lives and extend to many areas of society, including the industrialized world. A common alternative for controlling the movements and actions of the drone is through unwired tactile interfaces, for which different remote control devices are used. However, control through such devices is not a natural, human-like communication interface, which sometimes is difficult to master for some users. In this research, we experimented with a domain-based speech recognition architecture to effectively control an unmanned aerial vehicle such as a drone. The drone control was performed in a more natural, human-like way to communicate the instructions. Moreover, we implemented an algorithm for command interpretation using both Spanish and English languages, as well as to control the movements of the drone in a simulated domestic environment. We conducted experiments involving participants giving voice commands to the drone in both languages in order to compare the effectiveness of each, considering the mother tongue of the participants in the experiment. Additionally, different levels of distortion were applied to the voice commands to test the proposed approach when it encountered noisy input signals. The results obtained showed that the unmanned aerial vehicle was capable of interpreting user voice instructions. Speech-to-action recognition improved for both languages with phoneme matching in comparison to only using the cloud-based algorithm without domain-based instructions. Using raw audio inputs, the cloud-based approach achieves 74.81% and 97.04% accuracy for English and Spanish instructions, respectively. However, with our phoneme matching approach the results are improved, yielding 93.33% accuracy for English and 100.00% accuracy for Spanish.

Robots are extending their presence in domestic environments every day, it being more common to see them carrying out tasks in home scenarios. In the future, robots are expected to increasingly perform more complex tasks and, therefore, be able to acquire experience from different sources as quickly as possible. A plausible approach to address this issue is interactive feedback, where a trainer advises a learner on which actions should be taken from specific states to speed up the learning process. Moreover, deep reinforcement learning has been recently widely used in robotics to learn the environment and acquire new skills autonomously. However, an open issue when using deep reinforcement learning is the excessive time needed to learn a task from raw input images. In this work, we propose a deep reinforcement learning approach with interactive feedback to learn a domestic task in a Human–Robot scenario. We compare three different learning methods using a simulated robotic arm for the task of organizing different objects; the proposed methods are (i) deep reinforcement learning (DeepRL); (ii) interactive deep reinforcement learning using a previously trained artificial agent as an advisor (agent–IDeepRL); and (iii) interactive deep reinforcement learning using a human advisor (human–IDeepRL). We demonstrate that interactive approaches provide advantages for the learning process. The obtained results show that a learner agent, using either agent–IDeepRL or human–IDeepRL, completes the given task earlier and has fewer mistakes compared to the autonomous DeepRL approach.

Fire recognition from visual scenes is a demanding task due to the high variance of color and texture. In recent years, several fire-recognition approaches based on deep learning methods have been proposed to overcome this problem. However, building deep convolutional neural networks usually involves hundreds of layers and thousands of channels, thus requiring excessive computational cost, and a considerable amount of data. Therefore, applying deep networks in real-world scenarios remains an open challenge, especially when using devices with limitations in hardware and computing power, e.g., robots or mobile devices. To address this challenge, in this paper, we propose a lightweight and efficient octave convolutional neural network for fire recognition in visual scenes. Extensive experiments are conducted on FireSense, CairFire, FireNet, and FiSmo datasets. In overall, our architecture comprises fewer layers and fewer parameters in comparison with previously proposed architectures. Experimental results show that our model achieves higher accuracy recognition, in comparison to state-of-the-art methods, for all tested datasets.