Search

International audience; Collaborative robots have open new ways of designing assembly processes, thanks to their ability to share work space with operators. not only may they support the economical performance, but they can also improve the overall ergonomics. Building on existing work on task allocation problems, the authors study further the collaboration opportunities between operator and robot, namely cooperation phases (type of collaboration where both operator and robot act on the same work piece). This work proposes anewformulation of the related problem, and solutions are sought through heuristics methods, to investigate whether concurrent usage of different collaboration modes delivers better performance. The results indicate that cooperation mode enables higher process performances while controlling ergonomic risks. With a concern for real-life application, it has been applied on a real case study to verify its applicability.

International audience; This paper is a discussion on the question of ecological validity of virtual reality in the light of three studies that we have done in previous works. These works are chosen as a basis for the discussion because they are all designed to assess validity using one method : the comparison of user perception and behavior between real and virtual environements. The first study explores visual perception of complex materials , the second studies the role of visual feedback on user gestures and object manipulation, the third is a study of virtual reality as a tool for assessing the acceptability of human robot collaboration in a car factory. We discuss our methodology, the limits of validity of VR in our three use cases and suggest future developments in VR to provide design tools for more valid VR environments.

In the latest years, robotic technologies have been increasingly introduced in rehabilitation with the main purpose of reducing the costs and speeding up the recovery process of patients. However, most of the commercial devices impose a pre-programmed trajectory to the limbs of the patients, who therefore behave in a passive way. Another current major limitation is the inability to accurately evaluate the dynamics of the interaction between the patient and the robotic device. This interaction plays a central role in the mutual modulation of human and robot system behavior with respect of their standalone behavior. In particular, the prediction of the interaction can provide useful information to better design the exoskeleton as well as the rehabilitation treatment. This thesis presents my proposed solution for the development of a simulator able to dynamically simulate at the same time the actuated robot device, the human body, and their emerging physical interaction during the movement cooperation. The main idea behind this solution is to decompose the main system in different levels. I called the proposed solution Multi-Level modeling approach, which is the main topic of this thesis. I proposed the following decomposition: Human, Robot, and Boundary Level. The levels are integrated into a whole system in which each of them addresses specific challenges. The Human Level represents the subject who is wearing, for example, an exoskeleton for the lower limbs. To reach a symbiotic collaboration between the subject and the exoskeleton, the proposed approach has to include the subject's intentions and efforts. Moreover, user's internal transformations provide important information about the internal dynamic parameters modulation due to the external device. The Robot Level consists of the wearable robot system which supports the movements. Our proposed approach includes models of both device mechanics and control strategies. This allows to test different control strategies and find the one that better fits each specific patient's needs and characteristics. The last level is the Boundary Level, which has the main objective to model the human-robot mechanical power transfer, including also the non-idealities (such as dissipative forces), in order to accurately estimate interactions. Challenges emerged during the development of the simulator system were faced, investigating different solutions, and selecting and validating the most promising one. First, I selected a common software platform, able to simultaneously reproduce the dynamic behavior of the three levels. The common software platform allows to build a quite flexible system where different solution could be evaluated simply modifying model parameters. Among different available software, OpenSim was selected because it is well known and used for the dynamic study of human movement. Although OpenSim was well tested in biomechanics, it required a further evaluation as simulator for Robot and Boudary Levels. Performed tests and their motivations are reported in this work. Human internal dynamics parameters are modulated by the influence of the external device. I proposed to monitor this variation, taking into consideration the neural drive sent to the muscles. This can be done by measuring the muscles' electromyographic (EMG) signals, which are the electrical potential generated by muscle cells when they are activated, prior to muscle contraction. These signals can be used as input for a physiologically accurate human musculoskeletal model, to calculate the subject contribution to the movement. As the relation between EMGs and the generated muscle forces and joint moments is not linear, the neuromusculoskeletal model is indeed needed to replicate step-by-step all the internal transformations which occur from the excitation of the muscle to the joint movement. Estimation of the emerging interaction, during the human-robot cooperation, can be performed through an interaction model which is basically a set of contact models. Due to the specific rehabilitation purpose of our work, this contact model needs special attention. I introduced and validated a procedure to calibrate the contact models to improve the accuracy of the estimated interaction forces. One of the problems of using EMG signals is that, in order to acquire them in a non-invasive way, surface electrodes must be used; however, this means that the collected data quality is quite susceptible to the electrodes placements and decay, and to electric and magnetic interferences. In many contexts, such as home rehabilitation, this could be a limitation. An alternative solution to avoid the direct EMG measurement is presented in this work. The idea is that for some repetitive tasks, which are most interesting for rehabilitation, it is possible to substitute the direct data collection with a subject specific model of EMGs. The objective of this work is to provide an effective approach to estimate the emerging interaction during the human-robot movement cooperation. The Multi-Level Modeling approach, presented in this thesis, decomposes this complex problem allowing to find all the required components to realize a whole system able to reach this objective.

Η βιομηχανία παραγωγής ρούχων, ακόμα και σήμερα, αποτελεί ένα βιομηχανικό κλάδο όπου η αυτοματοποίηση είναι εν μέρει σχεδόν ανύπαρκτη και εν γένει σε εμβρυικό στάδιο. Ένα μεγάλο τμήμα της γραμμής παραγωγής αποτελεί το τμήμα κατά το οποίο κομμάτια υφάσματος ράβονται σε μία ραπτομηχανή. Το στάδιο της ραφής, παρόλο που είναι το μεγαλύτερο σε χρονική διάρκεια και το σημαντικότερο στην τελική ποιότητα, παραμένει σχεδόν εξολοκλήρου μία χειρονακτική εργασία. Οι ιδιαιτερότητες των υφασμάτων όπως: η πολύ μικρή αντίστασή τους σε κάμψη, οι μεγάλες παραμορφώσεις τους, η απρόβλεπτη στατική/δυναμική συμπεριφορά τους, η ανισότροπη και μη-γραμμική φύση τους και η ανομοιογένειά τους είναι οι κύριοι παράγοντες των έντονων δυσκολιών κατά τον χειρισμό τους. Η ρομποτική ραφή είναι ένα σχετικά νέο ερευνητικό πεδίο αλλά και εξαιρετικά πολύπλοκο πρόβλημα, όπου ελάχιστοι ερευνητές έχουν προσπαθήσει να προσεγγίσουν. Αυτή τη στιγμή, από όσο είναι δυνατόν να είναι γνωστόν, μόνο τρία ερευνητικά κέντρα ασχολούνται, σε διεθνή επίπεδο, ενώ το ένα από αυτά είναι η Ομάδα Ρομποτικής του Πανεπιστημίου Πατρών. Στόχος της διατριβής είναι η ανάπτυξη ενός ευέλικτου συστήματος το οποίο ενσωματώνει όλα εκείνα τα ευφυή χαρακτηριστικά ώστε να συμβάλει στη ρομποτική αυτοματοποίηση της ραφής υφασμάτων. Η επίτευξη αυτού του στόχου στηρίζεται σε μεθόδους υπολογιστικής νοημοσύνης, σε τακτικές εμπνευσμένες από τον τρόπο εργασίας του ίδιου του ανθρώπου και στην ποιοτική διαχείριση της γνώσης/δεδομένων που εμπεριέχουν ασάφεια. Ταυτόχρονα, επιτυγχάνεται ο έλεγχος της ραφής χωρίς τη χρήση αναλυτικών μοντέλων τόσο των υφασμάτων όσο και της διαδικασίας. Ο συντονισμός όλων των προηγουμένων επιτυγχάνεται από ένα ευέλικτο ευφυές ιεραρχικό σύστημα ελέγχου της διαδικασίας της ραφής. Στην παρούσα διατριβή αναπτύχθηκε ένα σύνολο νέων μεθόδων για την αυτοματοποιημένη ραφή υφασμάτων με ένα βιομηχανικό ρομπότ εφοδιασμένο με έναν αισθητήρα δύναμης και μία κλασσικού τύπου ραπτομηχανή. Η πρωτότυπη προσέγγιση που αποτελεί το πλαίσιο της διατριβής αποτελείται από ένα ιεραρχικό σύστημα εκτιμήσεων, αποφάσεων και ελέγχου της διαδικασίας ραφής. Γίνεται μία συστηματική μελέτη, καταγραφή και αξιολόγηση όλων των χειρισμών του υφάσματος, πριν και κατά τη διάρκεια της ραφής, ως προς τις απαιτήσεις τους σε συστήματα αισθητήρων αλλά και στρατηγικών ελέγχου. Αναπτύσσεται μία πρωτότυπη βάση δεδομένων και ένα έμπειρο σύστημα που ενσωματώνουν όλη αυτή την προγενέστερη γνώση και εμπειρία. Παρουσιάζεται η ιδέα της εφαρμογής μεθόδων υπολογιστικής νοημοσύνης στο επιστημονικό πεδίο της ραφής υφασμάτων και εφαρμογής μεθόδων ευφυούς ελέγχου, για την τάνυση των υφασμάτων, βασισμένες σε νευρωνικά δίκτυα οι οποίες, για πρώτη φορά, χρησιμοποιούν ποιοτικές εκτιμήσεις των ιδιοτήτων και των χαρακτηριστικών των υφασμάτων. Έτσι, εισάγεται η έννοια των λεκτικών μεταβλητών για την ποιοτική περιγραφή των ιδιοτήτων των υφασμάτων. Στο πλαίσιο αυτό, εκφράζεται για πρώτη φορά, ποιοτικά σαν ποσοστό, η έννοια της “εκτατότητας” (extensibility) των υφασμάτων, που είναι μία βασική και καθοριστική ιδιότητά τους κατά την εργασία της ραφής τους. Επίσης περιγράφεται η ανάπτυξη μιας νέας μεθόδου αυτοματοποιημένου πειραματικού προσδιορισμού της “εκτατότητας” των υφασμάτων. Η πρωτότυπη αυτή προσέγγιση εφαρμόστηκε σε μονά και διπλά υφάσματα ενώ τα πειράματα εφελκυσμού ενσωματώθηκαν “έξυπνα” στην ίδια τη ραπτομηχανή και λίγο πριν τη διαδικασία της ραφής. Παρουσιάζεται ένα ασαφές σύστημα εξαγωγής της επιθυμητής τάνυσης, κατά τη ραφή του κάθε υφάσματος, το οποίο βασίζεται στην ποιοτική εκτίμηση της “εκτατότητας” του υφάσματος και διαμορφώθηκε από την εμπειρία του έμπειρου χειριστή και την ασαφή συνεπαγωγή που κάνει αυτός ανάμεσα στην “εκτατότητα” και την επιθυμητή τάνυση. Αναπτύχθηκε μία νέα μέθοδος ελέγχου της τάνυσης του υφάσματος κατά τη ραφή του. Σε αυτή χρησιμοποιήθηκε ένας νευρωνικός ελεγκτής ευθείας τροφοδότησης για τον έλεγχο του ρομποτικού άκρου εργασίας με σκοπό τη διατήρηση της τάνυσης του υφάσματος σύμφωνα με την εκάστοτε επιθυμητή. Επιπροσθέτως, παρουσιάζεται η επέκταση της προτεινόμενης προσέγγισης, του νευρωνικού ελέγχου της τάνυσης, η οποία έδειξε τη δυνατότητα γενίκευσής της και σε άλλες παραπλήσιες εργασίες. Επίσης, καταγράφονται τα νέα ερευνητικά πεδία που ανοίγουν στο χώρο της συνεργασίας ανθρώπου-ρομπότ για τον χειρισμό υπερ-εύκαμπτων και εύκαμπτων αντικειμένων. Τελικά, παρουσιάζεται μία συστηματική εργαστηριακή εκτέλεση ραφών, σε μία μεγάλη ποικιλία τόσο μονών όσο και διπλών υφασμάτων, με ρομπότ ενώ συγκρίθηκαν αυτές με ραφές από έμπειρους. Μέσα από τις πειραματικές ραφές αναδεικνύεται, η αναζητούμενη ευελιξία του συστήματος και η ικανοποιητική απόκρισή του σε μία μεγάλη ποικιλομορφία υφασμάτων. The clothing industry, even today, is an industry where automation is partly almost nonexistent and generally in their infancy stages. A large part of the production line is the part in which pieces of fabrics are sewn using a sewing machine. The sewing process, while being the largest in duration and the most important to the final quality of the cloths, remains almost entirely one handiwork. The specifics of fabrics like: their very little bending resistance, their large deformations, their unpredictable static / dynamic behavior, their anisotropic and non-linear nature and heterogeneity are the main factors of acute difficulties in their handling. The robotized sewing of fabrics is a relatively new research field but it is also an extremely complex problem, where few researchers have tried to investigate it. Right now, as far as can be known, only three research centers in international level are involved this field, while one of them is the Robotics Group of the University of Patras. The aim of this thesis is to develop a flexible system which incorporates all those intelligent features in order to help automate robotic sewing of fabrics. This objective is based on computational intelligence methods, in approaches inspired by the way a human works and finally in the qualitative management of the knowledge/data that involve uncertainties. Simultaneously, the sewing control is achieved without using analytical models of both the fabrics and process. The coordination, of all of the previous mentioned, is achieved by a flexible hierarchical intelligent control system of the sewing process. This thesis developed a set of new set of methods for automated sewing of fabrics using an industrial robot, equipped with a force sensor, and a conventional sewing machine. The novel approach, which composes the main framework of this thesis, consists of a hierarchical system of estimations, decisions and process control of the sewing task. It is a systematic study, of recording and evaluating all the fabric handling tasks before and during the sewing process, according to the necessary requirements in sensor systems and control strategies. An original database and an expert system incorporating all this prior knowledge and experience is developed. The idea of applying methods of computational intelligence to the scientific field of sewing fabrics is presented. Also, the application of intelligent control methods, for stretching the fabrics during the sewing, which is based on neural networks while using qualitative assessment of the properties and characteristics of fabrics, is presented. Therefore, the concept of using linguistic variables for the qualitative description of the properties of fabrics, is introduced. In this context, it is the first time where the concept of the "extensibility" of a fabric is expressed quantitatively as a percentage. This is an essential and determining property of a fabric that should be taken into account during the sewing process. Also, the development of a new method for the automated experimental determination of "extensibility" of fabrics is described. This original approach has been applied to single and double layers of fabrics, while the tensile experiments are ‘intelligently’ incorporated into the same sewing machine and just before the sewing process. A fuzzy system for determining the desired tension for each of the fabrics is presented. This system is based on a qualitative assessment of "extensibility" of each fabric and it is initialized by the experience of the experts and the fuzzy implication they make between "extensibility" and desired tension. A new method to control the tension of the fabric during the sewing is developed. A feedforward neural controller, to control the robotic gripper to maintain the tension of the fabric according to the respective desired, has been used. Additionally, an expansion of the proposed approach is presented, for the case of the neuro-control of the tension, which showed the controller’s ability of generalization to other similar tasks. Also, new research areas that have been opened in the field of human-robot cooperation for handling non-rigid and very flexible objects are stated. Finally, there is presented a systematic experimental robotic execution of seams in a wide variety of both single and double layers of fabrics, while the, produced by the robot, seams are compared with the seams produced by the experts. Through experimental phase and the results of the stitched fabrics it is highlighted the desired flexibility of the system and its satisfactory response in a wide variety of fabrics.

Due to copyright restrictions, the access to the full text of this article is only available via subscription. We offer a system for a human-robot cooperation with natural communication in order to make using of robots in human-interactive tasks easier and more effective. This system includes a system with 3D motion capture cameras, and a motion recognition using Hidden Markov Models (HMM). We implemented the solution on a simulation using real data collected for this experiment.

This thesis explores the problem of mobile robot navigation in dense human crowds. We begin by considering a fundamental impediment to classical motion planning algorithms called the freezing robot problem: once the environment surpasses a certain level of complexity, the planner decides that all forward paths are unsafe, and the robot freezes in place (or performs unnecessary maneuvers) to avoid collisions. Since a feasible path typically exists, this behavior is suboptimal. Existing approaches have focused on reducing predictive uncertainty by employing higher fidelity individual dynamics models or heuristically limiting the individual predictive covariance to prevent overcautious navigation. We demonstrate that both the individual prediction and the individual predictive uncertainty have little to do with this undesirable navigation behavior. Additionally, we provide evidence that dynamic agents are able to navigate in dense crowds by engaging in joint collision avoidance, cooperatively making room to create feasible trajectories. We accordingly develop interacting Gaussian processes, a prediction density that captures cooperative collision avoidance, and a "multiple goal" extension that models the goal driven nature of human decision making. Navigation naturally emerges as a statistic of this distribution. Most importantly, we empirically validate our models in the Chandler dining hall at Caltech during peak hours, and in the process, carry out the first extensive quantitative study of robot navigation in dense human crowds (collecting data on 488 runs). The multiple goal interacting Gaussian processes algorithm performs comparably with human teleoperators in crowd densities nearing 1 person/m2, while a state of the art noncooperative planner exhibits unsafe behavior more than 3 times as often as the multiple goal extension, and twice as often as the basic interacting Gaussian process approach. Furthermore, a reactive planner based on the widely used dynamic window approach proves insufficient for crowd densities above 0.55 people/m2. We also show that our noncooperative planner or our reactive planner capture the salient characteristics of nearly any dynamic navigation algorithm. For inclusive validation purposes, we show that either our non-interacting planner or our reactive planner captures the salient characteristics of nearly any existing dynamic navigation algorithm. Based on these experimental results and theoretical observations, we conclude that a cooperation model is critical for safe and efficient robot navigation in dense human crowds. Finally, we produce a large database of ground truth pedestrian crowd data. We make this ground truth database publicly available for further scientific study of crowd prediction models, learning from demonstration algorithms, and human robot interaction models in general.

Zur Zeit haben Industrieroboter nur eine sehr begrenzte Wahrnehmung ihrer Umwelt. Wenn sich Menschen im Arbeitsraum des Roboters aufhalten sind sie daher gefährdet. Durch eine Einteilung der möglichen Roboterbewegung in verschiedene Klassen kann gezeigt werden, dass die für einen Menschen im Arbeitsraum gefährlichste Bewegung die freie Transferbewegung ist. Daher besteht die betrachtete Aufgabe darin, diese Transferbewegung eines Manipulators durchzuführen, ohne mit dynamischen Hindernissen, wie zum Beispiel Menschen, zu kollidieren. Das SIMERO-System gliedert sich in die vier Hauptkomponenten Bildverarbeitung, Robotermodellierung, Kollisionserkennung und Bahnplanung. Diese Komponenten werden im einzelnen vorgestellt. Die Leistungsfähigkeit des Systems und die weiteren Verbesserungen werden an einem Versuch exemplarisch gezeigt.

Zur Zeit haben Industrieroboter nur eine sehr begrenzte Wahrnehmung ihrer Umwelt. Wenn sich Menschen im Arbeitsraum des Roboters aufhalten sind sie daher gefährdet. Durch eine Einteilung der möglichen Roboterbewegung in verschiedene Klassen kann gezeigt werden, dass die für einen Menschen im Arbeitsraum gefährlichste Bewegung die freie Transferbewegung ist. Daher besteht die betrachtete Aufgabe darin, diese Transferbewegung eines Manipulators durchzuführen, ohne mit dynamischen Hindernissen, wie zum Beispiel Menschen, zu kollidieren. Das SIMERO-System gliedert sich in die vier Hauptkomponenten Bildverarbeitung, Robotermodellierung, Kollisionserkennung und Bahnplanung. Diese Komponenten werden im einzelnen vorgestellt. Die Leistungsfähigkeit des Systems und die weiteren Verbesserungen werden an einem Versuch exemplarisch gezeigt.

We present a system concept allowing humans to work safely in the same environment as a robot manipulator. Several cameras survey the common workspace. A look-up-table-based fusion algorithm is used to back-project directly from the image spaces of the cameras to the manipulator?s con-figuration space. In the look-up-tables both, the camera calibration and the robot geometry are implicitly encoded. For experiments, a conven-tional 6 axis industrial manipulator is used. The work space is surveyed by four grayscale cameras. Due to the limits of present robot controllers, the computationally expensive parts of the system are executed on a server PC that communicates with the robot controller via Ethernet.

Small and Medium sized Enterprises (SMEs) in Europe struggle to incorporate industrial robots in their production environments, while large enterprises use these robots for large batch production only. The paradigm shift from mass production to mass personalization decreases batch sizes and changes the approach to implementation of industrial robots in manufacturing environments. It also opens doors for SMEs to further incorporate robots in their production environments. The goal of this research is to evaluate the suitability of a data-centric, distributed, decentralized manufacturing system for cooperation between robots and humans. A case is presented featuring cooperation between robots and humans. A control system is proposed based on distributed intelligence and decentralized control, to handle the rapidly expanding complexity in dynamic manufacturing environments. The communication in such a distributed environment is provided by a Data Distribution Service system; an extendible, flexible approach to communication. Key issues that are encountered in implementing the cooperation into the current industrial environments are identified. The proposed control system is projected on the case and evaluated for application suitability and expected performance.