Abstract
The number of applications in which industrial robots share their working environment with people is increasing. Robots appropriate for such applications are equipped with safety systems according to ISO/TS 15066:2016 and are often referred to as collaborative robots (cobots). Due to the nature of human-robot collaboration, the working environment of cobots is subjected to unforeseeable modifications caused by people. Vision systems are often used to increase the adaptability of cobots, but they usually require knowledge of the objects to be manipulated. The application of machine learning techniques can increase the flexibility by enabling the control system of a cobot to continuously learn and adapt to unexpected changes in the working environment. In this paper we address this issue by investigating the use of Reinforcement Learning (RL) to control a cobot to perform pick-and-place tasks. We present the implementation of a control system that can adapt to changes in position and enables a cobot to grasp objects which were not part of the training. Our proposed system uses deep Q-learning to process color and depth images and generates an ?-greedy policy to define robot actions. The Q-values are estimated using Convolution Neural Networks (CNNs) based on pre-trained models for feature extraction. To reduce training time, we implement a simulation environment to first train the RL agent, then we apply the resulting system on a real cobot. System performance is compared when using the pre-trained CNN models ResNext, DenseNet, MobileNet, and MNASNet. Simulation and experimental results validate the proposed approach and show that our system reaches a grasping success rate of 89.9% when manipulating a never-seen object operating with the pre-trained CNN model MobileNet.

Abstract
Future power systems encourage the use of renewable energy resources, among them wind power is of great interest, but its power output is intermittent in nature which can affect the stability of the power system and increase the risk of blackouts. Therefore, a forecasting model of the wind speed is essential for the optimal operation of a power supply with an important share of wind energy conversion systems. In this paper, two wind speed forecasting models based on multiple meteorological measurements of wind speed and temperature are proposed and compared according to their mean squared error (MSE) value. The first model concerns the artificial intelligence based on neural network (ANN) where several network configurations are proposed to achieve the most suitable structure of the problem, while the other model concerned the Adaptive Neuro-Fuzzy Inference System (ANFIS). To enhance the results accuracy, the invalid input samples are filtered. According to the computational results of the two models, the ANFIS has delivered more accurate outputs characterized by a reduced mean squared error value compared to the ANN-based model.

Abstract
The proposed work was developed as part of the SmartHealth project, which aims to advance upper body rehabilitation by granting a robotic alternative to reduce the limitations of physical therapy while conferring more intensive and personalized therapy sessions for patients. The use of robots permits therapists to be relieved of laborious and repetitive tasks while supplying feedback for patients and physiotherapists through automatic recordings. The proposed strategy is to develop new python-based software that controls the robot, collects the patient's forces and muscle activity in real-time, and stores them for future analysis while providing visual feedback, thus allowing session optimization. These features permit the physiotherapist to objectively perceive the patient's performance during exercise. This solution is implemented in robots already commercialized in the industrial field. These kinds of robots are generally mass-produced in production lines at a relatively low cost and with great flexibility.

Abstract
In view of the actual climate change scenario felt across the globe, resource management is crucial, especially with regard to water. In this sense, continuous monitoring of plant water status is essential to optimise not only crop management but also water resources. Currently, monitoring of vine water status is done through expensive and time-consuming methods that do not allow continuous monitoring, which is especially inconvenient in places with difficult access. The aim of the developed work was to install three groups of sensors (Environmental, Plant and Soil) in a vineyard and connect them through LoRaWAN protocol for data transmission. The results demonstrate that the implemented system is capable of continuous data communication without data loss. The reduced cost and superior range of LoRaWAN compared to WiFi or Bluetooth is especially important for applications in remote areas where cellular networks have little coverage. Altogether, this methodology provides a remote, continuous and more effective method to monitor plant water status and is capable of supporting producers in more efficient management of their farms and water resources.

Abstract
Rehabilitation robotics aims to facilitate the rehabilitation procedure for patients and physical therapists. This field has a relatively long history dating back to the 1990s; however, their implementation and the standardisation of their application in the medical field does not follow the same pace, mainly due to their complexity of reproduction and the need for their approval by the authorities. This paper aims to describe architecture that can be applied to industrial robots and promote their application in healthcare ecosystems. The control of the robotic arm is performed using the software called SmartHealth, offering a 2 Degree of Autonomy (DOA). Data are gathered through electromyography (EMG) and force sensors at a frequency of 45 Hz. It also proves the capabilities of such small robots in performing such medical procedures. Four exercises focused on shoulder rehabilitation (passive, restricted active-assisted, free active-assisted and Activities of Daily Living (ADL)) were carried out and confirmed the viability of the proposed architecture and the potential of small robots (i.e., the UR3) in rehabilitation procedure accomplishment. This robot can perform the majority of the default exercises in addition to ADLs but, nevertheless, their limits were also uncovered, mainly due to their limited Range of Motion (ROM) and cost.

Abstract
Unmanned Aerial Systems (UAS) are becoming more attractive in diverse applications due to their efficiency in performing tasks with a reduced time execution, covering a larger area, and lowering human risks at harmful tasks. In the context of Oil & Gas (O&G), the scenario is even more attractive for the application of UAS for inspection activities due to the large extension of these facilities and the operational risks involved in the processes. Many authors proposed solutions to detect gas leaks regarding the onshore unburied pipeline structures. However, only a few addressed the navigation and tracking problem for the autonomous navigation of UAS over these structures. Most proposed solutions rely on traditional computer vision strategies for tracking. As a drawback, depending on lighting conditions, the obtained path line may be inaccurate, making a strategy to force the UAS to continue on the path necessary. Therefore, this research describes the potential of an autonomous UAS based on image processing technique and Convolutional Neural Network (CNN) strategy to navigate appropriately in complex unburied pipeline networks contributing to the monitoring procedure of the Oil & Gas Industry structures. A CNN is used to detect the pipe, while image processing techniques such as Canny edge detection and Hough Transform are used to detect the pipe line reference, which is used by a line following algorithm to guide the UAS along the pipe. The framework is assessed by a PX4 flight controller Software-in-The-Loop (SITL) simulations performed with the Robot Operating System (ROS) along with the Gazebo platform to simulate the proposed operational environment and verify the approach's functionality as a proof of concept. Real tests were also conducted. The results showed that the solution is robust and feasible to deploy in this proposed task, achieving 72% of mean average precision on detecting different types of pipes and 0.0111 m of mean squared error on the path following with a drone 2 m away from a tube.