Speech Emotion Recognition (SER) is a vital component of affective computing, enabling systems to interpret users’ emotional states through vocal cues. This paper presents the development of a real-time SER model based on a Long Short-Term Memory (LSTM) neural network, optimized for execution on embedded platforms such as the Raspberry Pi. The model is trained using two publicly available emotional speech datasets, RAVDESS and TESS, which provide diverse expressions of emotion in English. Their combination increases variability in vocal patterns and improves generalization to real-world conditions. The proposed system focuses on low-latency inference and efficient memory usage, enabling the classification of seven emotional categories within the computational constraints typical of embedded hardware. Performace is critical as it will be integrated into a multimodal emotion recognition system to be deployed in embedded systems. Experimental evaluations show that the LSTM-based architecture offers a strong balance between accuracy and efficiency, supporting its integration into real-time, demonstrating the feasibility of deploying emotionally aware systems on low-power devices.

Epilepsy is a chronic brain disease characterized by recurrent and transient seizures, which is accompanied by super-synchronous abnormal discharge of electroencephalogram (EEG) signals. As a non-invasive auxiliary diagnostic technique, EEG is currently important means of seizure detection. However, due to ignoring the spatial topological relationship between electrodes, existing data-driver methods fail to fully reflect the interaction between signals. Meanwhile, their models usually be designed with a large number of redundant parameters, making it difficult to deploy to micro-embedded devices with limited-resources. In this paper, we propose a on-device learning with edge device for epilepsy EEG classification network based on GCN (oDLGCN-EEG). Specifically, to analyze the spatial relationships between various electrodes and their temporal dependencies, we design a Brain Topology Network (BTN) for the spatiotemporal dependency feature map construction. To capture the internal activity during epilepsy seizures, we design a Neural Feature Extraction Module (NFEM) for the neural activity feature map construction. Besides, we propose a pruning scheme to optimize the model, which successfully deploys the optimized oDLGCN-EEG on the embedded Raspberry Pi device for efficient and low-power consumption intelligent epilepsy classification. Experiments in comparison with state-of-the-art methods show that oDLGCN-EEG achieves the best classification accuracy and with the smallest parameter number on baseline EEG dataset. The code is available at https://github.com/cathnat/Epileptic_Classification.

In this work a set of novel approaches based on well-known computer vision techniques is proposed to deal with the autonomous perception part of an HRI robotic system. In the considered scenario, a human-robot duo interact to plan maintenance and pruning operations in a vineyard. Once the human expert has selected a set of branches to cut in the approximate area, the robot stores this information in order to be able to return later, identify the branches and cut them at an adequate point. This is achieved by producing models of the plant by segmenting the view through a combination of intensity and depth data, using RGB-D camera sensors. This segmentation is fed back into the pipeline, used as a base mask to identify parts of the plant through watershed segmentation, and identifying the branches and gems in the segmented images. The proposed method was develop based on real data, and tested in experimental scenarios with the real robot.

Pipelines are used for fluid transportation over long distances, necessitating effective monitoring and maintenance strategies. Pipeline Inspection Gauges (PIGs) instrumented with sensors and embedded systems playa crucial role in this field. This Work in Progress explores using a PIG for trajectory estimation and data acquisition. Several sensing technologies, including Inertial Navigation Systems (INS) and odometers, are integrated to determine the location and parameters of the pipeline. The electronic architecture of the PIG, designed for trajectory estimation, is discussed, focusing on the utilization of odometers and an Inertial Measurement Unit (IMU). The system features an FPGA-based electronic acquisition system, incorporating Quadrature Pulse Modules and Counters for data storage and processing. Furthermore, preliminary tests conducted on a pipeline circuit demonstrate the effectiveness of the proposed electronic architecture. Overall, this study presents a comprehensive approach to pipeline trajectory estimation, emphasizing the importance of data acquisition and processing to ensure accurate pipeline location.

This work presents a solution to automatize sampling tasks in a wastewater treatment plant with open air basins. At the behest of human operators, a set of UAVs managed as a network of autonomous agents will perform sample missions by taking direct measurements (through a multiparametric probe) or capturing samples and carrying them to the laboratory with specific developed hardware. These capabilities allow the proposed solution to act as a virtual sensor network with sampling points deployed and connected at any point reachable by UAVs. The hardware prototypes are fully described, with focus on the integration of systems, and the software architecture used is analysed and fully justified. Special focus was put on the localization problem, and several solutions were evaluated. Experimental results of the prototype UAV and sampling probe built are provided to validate the hardware designs, with focus on the localizations tasks.

One of the most critical issues that arises when controlling a robot is the necessity to locate itself in the environment. Ranging from industrial processes applications to outdoor mobile robots, landmarks are used to such purpose: by recognizing them, the robots are able to know where they are with the aid of computer vision. Fiducial markers are the cheapest and one of the most common solutions to this issue: 2D planar patterns that embed some information that can be identified using artificial vision techniques. Many different typologies of markers as well as image processing algorithms are being implemented nowadays, using C++ / Python / MATLAB® libraries and ROS as middleware. In this work we have evaluated the robustness of various fiducial marker typologies, documenting the main technical aspects of the used implementations and commenting how each algorithm operates. Aggregated results are presented and discussed, evaluating the different implementations tested.

This work presents a solution to automatize the water sampling process of outdoor basins in a wastewater treatment plant. The system proposed is based on the utilization of collaborative robotics: a team of an UAV and a terrestrial robotic platform make a route along the plant collecting and storing the water samples. The architecture of the designed system is described in terms of functional blocks, and implementation details including software frameworks and hardware on the UAV are provided. As the objective of the system is industry levels of robustness and performance, the UAV use is minimized and subjected to control from the robotic ground platform, reducing risks associated with autonomous UAV. To conclude, results from experiments performed to validate the viability of the system and study several design decisions are presented and briefly discussed, including: estimation of the accuracy of several GNSS technologies on the plant, viability of the landing operation over a mobile robotic platform and controlling a quadrotor over waters.

A complete approach to the visual localization and mapping problem (SLAM) is presented in this work. The presented approach exploits the enhanced capabilities of a system where a human and a robot collaborate in surveying/exploratory tasks. The human is supposed to wear a smart headwear device, which deploys a inertial measurement unit and a camera, Hv. This camera acts as a secondary sensor, and provides data to the robotic Rvcamera performing mapping tasks. The data from the human-worn camera is used to produce real-time depth estimation of landmarks when its field of view overlaps with that of Rv. These measurements are mathematically fully integrated into the EKF-SLAM methodology. Experiments with real captured data validate the proposed approach.

A novel approach to the SLAM problem has been tested in an industrial environment within a robotic assistance context. In order to be fully reliable in non-modelled circumstances where the environment cannot be considered as known a priori, a robot assistant must be able to localize and map its environment. The use of a camera sensor to solve localization has several advantages and weaknesses due the nature of the only-bearing data. But as the robot is expected to assist the human agent, this agent can deploy additional sensors and provide the robot with data to help solve the SLAM problem. Thus, another camera worn by the human agent is used to produce non-continuous stereo data with the robotic camera, to speed-up and add robustness to several parts of the monocular SLAM process considered. The approach has been tested with real experiments focused on singular trajectories and other issues found on industrial environments.

This work presents a complete navigation architecture for an autonomous aerial robot. The proposed scheme consist of: i) a low-level controller for establishing the attitude and position of the vehicle, ii) a Simultaneous Localization and Mapping (SLAM) system, based in bearing (angular) measurements, which gives the robot the ability for navigating in unknown environments, and iii) a high-level motion control system which generates online trajectories. The high-level motion control system (MCS), which represents the main contribution of this work, is inspired by the behavior-based control strategies. The MCS takes as input a very high level mission target (e.g. “explore as much as you can”) and generates online trajectories according to the mission, but at the same time minimizing uncertainty in the estimations in order to maintain the integrity of the robot. The proposed architecture is explained for simplified 3DOF dynamics, but it could be extended in a straightforward manner in order to be applied to full dynamics. Several simulations are included in order to show the performance of the proposed scheme.

This paper describes a new approach to the well-known robotics problem of simultaneous location and mapping (SLAM). The proposed technique introduces a linear varying parameter (LPV) modeling solution for the estimation of nonlinear models in a Kalman Filter based algorithm. In this technique, the estimation model for the robotic device considered is modeled as a quasi-LPV model, which in turn, is linearized around a set of given points of the varying parameter. The observation model is rearranged into a pseudo-measurement model, which is used in form of a pseudo-linear model during the update stage of the Kalman filter. The initial tests and experimentations suggest that this technique can improve Extended Kalman Filter SLAM results by avoiding a great deal of the bias introduced by linearization of nonlinear models into EKF equations.

In this paper a novel Simultaneous Localization and Mapping (SLAM) is presented. Using the sound as the input signal, instead of the classical vision or laser systems, leads to a SSLAM (sound SLAM) with a new features, such as the use of a Linear Parameter Varying (LPV) Kalman filter rather than the classical Extended Kalman filter. The other novelty is the modeling of sound reverberation using LPV models. The work is an extension under development from previous research group's works. The experimental partial results and the theoretical developments encourage authors to follow this unexplored line of SSLAM.

Non-speech audio is a non-explored characteristic in robot localization but due to its potentiality it can yield a valuable information together with other sensorial systems. In this work, a novel robot localization method is proposed based on audio signal pattern recognition with extracted features from signal identification. To reinforce the localization, avoiding ambiguity and reducing uncertainty, a sensorial system is used aboard the robot to compute the angle between itself and the sound source. This method can be generalized to any non-speech sound signal because the acoustical meaning and the room geometry are related.