This review article discusses the definition and implementation of brain–computer interface (BCI) system relying on brain connectivity (BC) and machine learning/deep learning (DL) for motor imagery (MI)-based applications. During the past few years, many approaches have been explored in terms of types of neurological sources of information, feature extraction, and intention prediction for BCI applications. Two novel aspects are becoming increasingly interesting for the BCI community: BC modeling and DL. The former aims at describing the interactions among different brain regions as connectivity patterns that reflect the dynamics of information flow either at rest or when performing a task. The latter is becoming pervasive for its capability of modeling and predicting complex data, where a huge amount of information is involved. In this scenario, we conducted a systematic literature review on BCI studies that led to the selection of 34 articles meeting all the required criteria. This provides evidence of the rapid growth of the topic over the past few years, though being still in its infancy. The last part of this article is dedicated to this new frontier of BCI that we call MI BC-based computer interfaces highlighting the potential of BC features. This, jointly with DL as enabling technology, has the potential of improving the performance of electroencephalography-based systems.

This position paper discusses the definition and implementation of a Brain Computer Interface (BCI) system based on Brain Connectivity (BC) and Machine Learning (ML) in motor imagery applications. BCI has become a well-established technique to control external devices and consequently help people in their everyday life. During the years many approaches have been explored in terms of neurological information, feature extraction, signal processing, and intention prediction. Two novel aspects are becoming increasingly interesting for the BCI community, i.e. BC modeling and ML techniques. The former aims at describing the interactions among different brain regions as connectivity patterns that reflect the dynamics of information flow at rest or when performing a task. The latter is becoming pervasive for its capability of modeling and predicting complex scenarios where a huge amount of data is involved. These aspects are relevant when considered by themselves, but become crucial assets when combined together. In this scenario, our research idea consists in identifying how the information in the two domains can be merged in a whole entity under a theoretical point of view: BC and ML applied to Electroencephalography (EEG) signals could provide the generalization capability, high accuracy and minimal computational time that allow researchers to build a reliable BCI capable to operate online.

The robotic industry needs new, innovative, ideas to be globally competitive. Conventional industrial robots are not able to adapt to changes in the assembly processes. Flexible assembly applications are actually uncommon and only isolated attempts exploit industrial robots to perform tasks with variability in the parts. Variability aspects are emphasized when developing novel manufacturing applications involving human–robot collaboration which are the foundation of Industry 4.0 systems. In this chapter, we describe how variability can be considered and mathematically described as part of the problem to obtain a flexible robotic solution. The selected approach is based on a probabilistic representation of the task obtained starting from a set of demonstrations collected from humans. The chapter illustrates the different steps leading to the complete learning framework. We start by describing the strategies adopted during the data collection phase. From the raw data, the design of feature extraction procedures is provided alongside a set of preprocessing techniques used to remove noisy and incoherent information. The resulting data set is used to train a model of task by following a probabilistic approach. The output of the model is exploited to actuate an industrial manipulator in the context of significant production scenarios. The robot motion strategies are also analyzed depending on the level of flexibility requested from the specific use case. Two main use cases are introduced: (1) the automatic assembly of a car door with its module, and (2) the robotized manufacturing process of electric machines, in particular winding of coils on stator or rotor cores. Each problem is mathematically formulated by modeling both the robotic platform and the target. The influence of the scenario variability with respect to the computed robotic motion is considered. The system flexibility is evaluated by means of an extensive set of benchmarking tests by recording data and actuating robots in both simulated and real environments. Achievements are compared with respect to state-of-the-art solutions by defining a set of objectives and metrics. The goal is to measure the performance of the system, for example, in minimizing the time and energy needed to move the robot in the working space, in generating an effective human–robot interaction with low reaction time and high accuracy, and in providing an intuitive robot learning technique to easily allow the human to teach the robot new tasks. Dynamic online reconfigurability of the framework is considered by testing its capability to deal with novel situations and new products. The integration of the proposed technologies with current robotic systems is discussed and a solution based on the robot operating system is proposed to provide a good infrastructure for network communication as well as all the tools necessary for a modern distributed and heterogeneous system. The feasibility and cost-effectiveness of the developed solutions are taken into account in order to demonstrate the applicability of the proposed approach in actual industrial settings.

In this paper, we propose a novel human-robot interface capable to anticipate the user intention while performing reaching movements on a working bench in order to plan the action of a collaborative robot. The system integrates two levels of prediction: motion intention prediction, to detect movements onset and offset; motion direction prediction, based on Gaussian Mixture Model (GMM) trained with IMU and EMG data following an evidence accumulation approach. Novel dynamic stopping criteria have been proposed to flexibly adjust the trade-off between early anticipation and accuracy. Results show that our system outperforms previous methods, achieving a real-time classification accuracy of 94.3±2.9% after 160.0msec±80.0msec from movement onset. The proposed interface can find many applications in the Industry 4.0 framework, where it is crucial for autonomous and collaborative robots to understand human movements as soon as possible to avoid accidents and injuries.

BACKGROUND:A proper modeling of human grasping and of hand movements is fundamental for robotics, prosthetics, physiology and rehabilitation. The taxonomies of hand grasps that have been proposed in scientific literature so far are based on qualitative analyses of the movements and thus they are usually not quantitatively justified.METHODS:This paper presents to the best of our knowledge the first quantitative taxonomy of hand grasps based on biomedical data measurements. The taxonomy is based on electromyography and kinematic data recorded from 40 healthy subjects performing 20 unique hand grasps. For each subject, a set of hierarchical trees are computed for several signal features. Afterwards, the trees are combined, first into modality-specific (i.e. muscular and kinematic) taxonomies of hand grasps and then into a general quantitative taxonomy of hand movements. The modality-specific taxonomies provide similar results despite describing different parameters of hand movements, one being muscular and the other kinematic.RESULTS:The general taxonomy merges the kinematic and muscular description into a comprehensive hierarchical structure. The obtained results clarify what has been proposed in the literature so far and they partially confirm the qualitative parameters used to create previous taxonomies of hand grasps. According to the results, hand movements can be divided into five movement categories defined based on the overall grasp shape, finger positioning and muscular activation. Part of the results appears qualitatively in accordance with previous results describing kinematic hand grasping synergies.CONCLUSIONS:The taxonomy of hand grasps proposed in this paper clarifies with quantitative measurements what has been proposed in the field on a qualitative basis, thus having a potential impact on several scientific fields.

Thanks to the increasing interest on robotics prosthetic devices controlled by means of physiological signals, a continuously increasing number of solutions are proposed. Usually the proposed solutions are very expensive and created ad-hoc for the final user. For this reason, a large part of the possible users can not afford this kind of technology. Furthermore, the software adaptation to the user is time consuming and physically stressing for the subject. The paper presents a low cost prosthesis framework, which covers the three fundamental aspects of a rehabilitation system, i.e. the prosthesis, the sensors used to record the physiological signals, and the software connecting the two previous points. To reduce the costs we chose a 3D printed prosthetic hand from an open-source project. We recorded Electromyography (EMG) signals from the subjects' muscles by using a low cost armband, a all-in-one solution easy to wear and remove. The EMG signals are preprocessed in order to be used online, and they are used to train a probabilistic model for classification purposes. Furthermore, the model is built on data from different subjects, in order to develop a subject-independent framework, which can be used by any subject, with no need of draining training phases. We test the goodness of our solution with a leave-one-out approach by classifying three different hand grasps. Finally, the 3D printed hand reproduces the movement performed by the subject. Data were recorded from four different subjects, each of them repeating the selected movements five times, and we obtained an overall accuracy of 76.8%.

European electrical machines manufacturers need to increase the flexibility of production process, due to the high cost of equipment setup at the beginning of each new production batch. Overall, most of these European manufacturers are striving to reduce costs while preserving the quality of products, in order to face the competition by Far East companies. There is a strong need for increasing productivity, flexibility and quality. In particular, in wound coils manufacturing process, current technologies allow only to big international manufacturer to automate their production lines, due to high machinery cost and set-up time, while small and medium manufacturers are forced to direct themselves towards manual production. This work aims to reduce costs and increase flexibility with the following contributions: (1) important reduction of setup time and costs of the winding machine, thanks to the simplicity and flexibility of the proposed approach; (2) increase in the quality of the final motors, thanks to the increased amount of copper that the robot will be able to insert in each coil with respect to manual winding; (3) possibility to parallelize the winding operations, dramatically increasing production rate; (4) decreased number of defected cores, thanks to an advanced quality inspection system; (5) reduction of environmental impact of the production process, thanks to a reduction of wasted copper wire.

Ahstract- The interest on wearable prosthetic devices has boost the research for a robust framework to help injured subjects to regain their lost functionality. A great number of solutions exploit physiological human signals, such as Electromyography (EMG), to naturally control the prosthesis, reproducing what happens in the human limbs. In this paper, we propose for the first time a way to integrate EMG signals with Inertial Measurement Unit (IMU) information, as a way to improve subject-independent models for controlling robotic hands. EMG data are very sensitive to both physical and physiological variations, and this is particularly true between different subjects. The introduction of IMUs aims at enriching the subject-independent model, making it more robust with information not strictly dependent from the physiological characteristics of the subject. We compare three different models: the first based on EMG solely, the second merging data from EMG and the 2 best IMUs available, and the third using EMG and IMUs information corresponding to the same 3 electrodes. The three techniques are tested on two different movements executed by 35 healthy subjects, by using a leave-one-out approach. The framework is able to estimate online the bending angles of the joints involved in the motion, obtaining an accuracy up to 0.8634. The resulting joint angles are used to actuate a robotic hand in a simulated environment.

In this paper, we address the problem of deploying a wire along a specific path selected by an unskilled user. The robot has to learn the selected path and pass a wire through the peg table by using the same tool. The main contribution regards the hybrid use of Cartesian positions provided by a learning procedure and joint positions obtained by inverse kinematics and motion planning. Some constraints are introduced to deal with non-rigid material without breaks or knots. We took into account a series of metrics to evaluate the robot learning capabilities, all of them over performed the targets.

The paper describes our experience in the benchmarking phase of the European Robotics Challenges project. The main focus is on the original solution proposed for solving a door assembly task. The proposal has to deal with tolerances in the door and module positions, never seen before doors, fast and usable human-machine interfaces, legacy hardware in industrial scenarios, and valuable results in benchmarking activities.

The interaction with robotic devices by means of physiological human signals has become of great interest in the last years because of the capability of catching human intention of movement and translate it in a coherent action performed by a robotic platform. Due to the complexity of EMG signals, several studies have been carried out about models built on a single subject (subject-specific). However, the execution of a certain task presents a common underlying behaviour, even if it is performed by different people. This common behaviour leads to some constraints that could be extracted by looking to different interpretations of the task, obtaining a subject-independent model. The few attempts in literature showed the possibility of creating a multiuser interface able to adapt to novel users (subject-independent). Nevertheless, the majority of the studies focused on classification problems, that are only able to determine the type of movement. We improved the state-of-the-art by introducing an online subject-independent framework able to compute the actual trajectory of the robot motion through a regression technique. The framework is based on a Gaussian Mixture Model (GMM) trained through Surface Electromyography (sEMG) signals coming from human subjects. Wavelet Transform has been used to elaborate the sEMG signals in real time. The goodness of the proposed framework has been tested with two different dataset involving various joints for both upper and lower limbs. The achieved results show that our framework could obtain high performances in both accuracy and computational time by reaching significant correlation (≥ 0.8). The whole procedure has been tested on two robots, a simulated hand and a humanoid, by remapping the human motion to the robotic platforms in order to verify the proper execution of the original movement.