Development of Autonomous Underwater Vehicles (AUVs) has permitted the automatization of many tasks originally achieved with manned vehicles in underwater environments. Teams of AUVs designed to work within a common mission are opening the possibilities for new and more complex applications. In underwater environments, communication, localization, and navigation of AUVs are considered challenges due to the impossibility of relying on radio communications and global positioning systems. For a long time, acoustic systems have been the main approach for solving these challenges. However, they present their own shortcomings, which are more relevant for AUV teams. As a result, researchers have explored different alternatives. To summarize and analyze these alternatives, a review of the literature is presented in this paper. Finally, a summary of collaborative AUV teams and missions is also included, with the aim of analyzing their applicability, advantages, and limitations.

In this paper, a mask region-based convolutional neural network (Mask R-CNN) is used to improve the performance of machine vision in the challenging task of detecting peduncles and fruits of green sweet peppers (Capsicum annuum L.) in greenhouses. One of the most complicated stages of the sweet pepper harvesting process is to achieve a precise cut of the peduncle or stem because this type of specialty crop cannot be grabbed and pulled by the fruit since the integrity and value of the product are compromised. Therefore, accurate peduncle detection becomes vital for the autonomous harvesting of sweet peppers. ResNet-101 combined with the feature pyramid network (FPN) architecture (ResNet-101 + FPN) is adopted as the backbone network for feature extraction and object representation enhancement at multiple scales. Mask images of fruits and peduncles are generated, focused on green sweet pepper, which is the most complex color variety due to its resemblance to the background. In addition to bounding boxes, Mask R-CNN provides binary masks as a result of instance segmentation, which would help improve the localization process in 3D space, the next phase of the autonomous harvesting process of sweet peppers, since it isolates the pixels belonging to the object and demarcates its boundaries. The prediction results of 1148 fruits on 100 test images showed a precision rate of 84.53%. The prediction results of 265 peduncles showed a precision rate of 71.78%. The mean average precision rate with an intersection over union at 50 percent (mAP@IoU=50) for model-wide instance segmentation was 72.64%. The average detection time for sweet pepper fruit and peduncle using high-resolution images was 1.18 s. The experimental results show that the proposed implementation manages to segment the peduncle and fruit of the green sweet pepper in real-time in an unmodified production environment under occlusion, overlap, and light variation conditions with effectiveness not previously reported for simultaneous 2D detection models of peduncles and fruits of green sweet pepper.

Mobile robots must be capable to obtain an accurate map of their surroundings to move within it. To detect different materials that might be undetectable to one sensor but not others it is necessary to construct at least a two-sensor fusion scheme. With this, it is possible to generate a 2D occupancy map in which glass obstacles are identified. An artificial neural network is used to fuse data from a tri-sensor (RealSense Stereo camera, 2D $360{}^{°}$ LiDAR, and Ultrasonic Sensors) setup capable of detecting glass and other materials typically found in indoor environments that may or may not be visible to traditional 2D LiDAR sensors, hence the expression improved LiDAR. A preprocessing scheme is implemented to filter all the outliers, project a 3D pointcloud to a 2D plane and adjust distance data. With a Neural Network as a data fusion algorithm, we integrate all the information into a single, more accurate distance-to-obstacle reading to finally generate a 2D Occupancy Grid Map (OGM) that considers all sensors information. The Robotis Turtlebot3 Waffle Pi robot is used as the experimental platform to conduct experiments given the different fusion strategies. Test results show that with such a fusion algorithm, it is possible to detect glass and other obstacles with an estimated root-mean-square error (RMSE) of 3 cm with multiple fusion strategies.

Several strategies to deal with the trajectory tracking problem of Unmanned Underwater Vehicles are encountered, from traditional controllers such as Proportional Integral Derivative (PID) or Lyapunov-based, to backstepping, sliding mode, and neural network approaches. However, most of them are model-based controllers where it is imperative to have an accurate knowledge of the vehicle hydrodynamic parameters. Despite some sliding mode and neural network-based controllers are reported as model-free, just a few of them consider a solution with finite-time convergence, which brings strong robustness and fast convergence compared with asymptotic or exponential solutions and it can also help to reduce the power consumption of the vehicle thrusters. This work aims to implement a model-free high-order sliding-mode controller and synthesize it with a time-base generator to achieve finite-time convergence. The time-base was included by parametrizing the control gain at the sliding surface. Numerical simulations validated the finite-time convergence of the controller for different time-bases even in the presence of high ocean currents. The performance of the obtained solution was also evaluated by the Root Mean Square (RMS) value of the control coefficients computed for the thrusters, as a parameter to measure the power consumption of the vehicle when following a trajectory. Computational results showed a reduction of up to 50% in the power consumption from the thrusters when compared with other solutions.

Autonomous mobile robots are an important focus of current research due to the advantages they bring to the industry, such as performing dangerous tasks with greater precision than humans. An autonomous mobile robot must be able to generate a collision-free trajectory while avoiding static and dynamic obstacles from the specified start location to the target location. Machine learning, a sub-field of artificial intelligence, is applied to create a Long Short-Term Memory (LSTM) neural network that is implemented and executed to allow a mobile robot to find the trajectory between two points and navigate while avoiding a dynamic obstacle. The input of the network is the distance between the mobile robot and the obstacles thrown by the LiDAR sensor, the desired target location, and the mobile robot’s location with respect to the odometry reference frame. Using the model to learn the mapping between input and output in the sample data, the linear and angular velocity of the mobile robot are obtained. The mobile robot and its dynamic environment are simulated in Gazebo, which is an open-source 3D robotics simulator. Gazebo can be synchronized with ROS (Robot Operating System). The computational experiments show that the network model can plan a safe navigation path in a dynamic environment. The best test accuracy obtained was 99.24%, where the model can generalize other trajectories for which it was not specifically trained within a $15 \mathrm{cm}$ radius of a trained destination position.

This study presents an approach to address the challenges of recognizing the maturity stage and counting sweet peppers of varying colors (green, yellow, orange, and red) within greenhouse environments. The methodology leverages the YOLOv5 model for real-time object detection, classification, and localization, coupled with the DeepSORT algorithm for efficient tracking. The system was successfully implemented to monitor sweet pepper production, and some challenges related to this environment, namely occlusions and the presence of leaves and branches, were effectively overcome. We evaluated our algorithm using real-world data collected in a sweet pepper greenhouse. A dataset comprising 1863 images was meticulously compiled to enhance the study, incorporating diverse sweet pepper varieties and maturity levels. Additionally, the study emphasized the role of confidence levels in object recognition, achieving a confidence level of 0.973. Furthermore, the DeepSORT algorithm was successfully applied for counting sweet peppers, demonstrating an accuracy level of 85.7% in two simulated environments under challenging conditions, such as varied lighting and inaccuracies in maturity level assessment.