Proceedings of the Fourth International Conference on Computer Vision Theory and Applications
We register close-range depth images of objects using a Swissranger sensor and apply a spring-mas... more We register close-range depth images of objects using a Swissranger sensor and apply a spring-mass model for 3D object reconstruction. The Swissranger sensor delivers depth images in real time which have, compared with other types of sensors, such as laser scanners, a lower resolution and are afflicted with larger uncertainties. To reduce noise and remove outliers in the data, we treat the point cloud as a system of interacting masses connected via elastic forces. We investigate two models, one with and one without a surface-topology preserving interaction strength. The algorithm is applied to synthetic and real Swissranger sensor data, demonstrating the feasibility of the approach. This method represents a preliminary step before fitting higher-level surface descriptors to the data, which will be required to define object-action complexes (OACS) for robot applications.
The ability to deal with articulated objects is very important for robots assisting humans. In th... more The ability to deal with articulated objects is very important for robots assisting humans. In this work, a framework to robustly and adaptively operate common doors, using an autonomous mobile manipulator, is proposed. To push forward the state-of-the-art in robustness and speed performance, we devise a novel algorithm that fuses a convolutional neural network with efficient point cloud processing. This advancement enables real-time grasping pose estimation for multiple handles from RGB-D images, providing a speed up improvement for assistive human-centered applications. In addition, we propose a versatile Bayesian framework that endows the robot with the ability to infer the door kinematic model from observations of its motion and learn from previous experiences or human demonstrations. Combining these algorithms with a Task Space Region motion planner, we achieve efficient door operation regardless of the kinematic model. We validate our framework with real-world experiments usin...
The expectation of a function of random variables is modelled as the value of the function in the... more The expectation of a function of random variables is modelled as the value of the function in the mean value of the variables plus a penalty term. This penalty term is calculated exactly, and the properties of different approximations are analyzed. In particular, for quadratic functions, the penalty term is shown to have a supervised and an unsupervised part. Then, two algorithms for minimizing the expected error of a feedforward network of random weights are compared. One of them is stochastical, while the other is deterministic and more flexible. Given a particular feedforward network architecture and a training set, the deterministic algorithm here presented accurately finds the weight configuration that makes the network response most resistant to a class of weight perturbations. Finally, the study of the most stable configurations of a network unravels some undesirable properties of networks with asymmetric activation functions
Es presenta un metode per estimar el moviment d’un robot a partir del seguiment de contorns actiu... more Es presenta un metode per estimar el moviment d’un robot a partir del seguiment de contorns actius en una successio d’imatges adquirides amb un sistema monocular i calibrat, muntat sobre el propi robot. Un cop el contorn actiu ha estat ajustat al voltant de la projeccio d’un objecte al pla de la imatge, les deformacions afins que pateix el contorn en la sequencia es parametritzen utilitzant un vector de forma de dimensio constant. S’analitzen les implicacions d’utilitzar vectors de forma de dimensio 3, 4 i 6. A partir d’aquests vectors de forma es pot trobar la rotacio i translacio 3D que ha provocat les deformacions, es a dir, el moviment propi de la camera. Els resultats preliminars obtinguts amb un robot rodant sobre un terra pla i vectors de forma de 4 components son forca encoratjadors.
In this work we propose a decision-making system that efficiently learns behaviors in the form of... more In this work we propose a decision-making system that efficiently learns behaviors in the form of rules using natural human instructions about cause-effect relations in currently observed situations, avoiding complicated instructions and explanations of long-run action sequences and complete world dynamics. The learned rules are represented in a way suitable to both reactive and deliberative approaches, which are thus smoothly integrated. Simple and repetitive tasks are resolved reactively, while complex tasks would be faced in a more deliberative manner using a planner module. Human interaction is only required if the system fails to obtain the expected results when applying a rule, or fails to resolve the task with the knowledge acquired so far. I. INTRODUCTION N this work we are facing the problem of decision making for a multitask robot embedded in a human environment that should rapidly learn to perform tasks by interacting with humans, in an on-line way, and without any previo...
This work presents an evaluation of three color constancy techniques applied to a landmark detect... more This work presents an evaluation of three color constancy techniques applied to a landmark detection system designed for a walking robot, which has to operate in unknown and unstructured outdoor environments. The first technique is the well-known image conversion to a chromaticity space, and the second technique is based on successive lighting intensity and illuminant color normalizations. Based on a differential model of color constancy, we propose the third technique, based on color ratios, which unifies the processes of color constancy and landmark detection. The approach used to detect potential landmarks, which is common to all evaluated systems, is based on visual saliency concepts using multiscale color opponent features to identify salient regions in the images. These regions are selected as landmark candidates, and they are further characterized by their features for identification and recognition.
Resumen: La cuestion de la decidibilidad, es decir, si es posible demostrar computacionalmente qu... more Resumen: La cuestion de la decidibilidad, es decir, si es posible demostrar computacionalmente que una expresion matematica es verdadera o falsa, fue planteada por Hilbert y permanecio abierta hasta que Turing la respondio de forma negativa. Establecida la no-decidibilidad de las matematicas, los esfuerzos en informatica teorica se centraron en el estudio de la complejidad computacional de los problemas decidibles. En este articulo presentamos una breve introduccion a las clases P (problemas resolubles en tiempo polinomico) y nP (problemas resolubles de manera no determinista en tiempo polinomico), al tiempo que exponemos la dificultad de establecer si P = nP y las consecuencias que se derivarian de que ambas clases de problemas fueran iguales. Esta cuestion tiene implicaciones no solo en los campos de la informatica, las matematicas y la fisica, sino tambien para la biologia, la sociologia y la economia. La idea seminal del estudio de la complejidad computacional es consecuencia di...
Along with its potential contributions to the practice of care, social assistive robotics raises ... more Along with its potential contributions to the practice of care, social assistive robotics raises significant ethical issues. The growing development of this technoscientific field of intelligent robotics has thus triggered a widespread proliferation of ethical attention towards its disruptive potential. However, the current landscape of ethical debate is fragmented and conceptually disordered, endangering ethics' practical strength for normatively addressing these challenges. This paper presents a critical literature review of the ethical issues of social assistive robotics, which provides a comprehensive and intelligible overview of the current ethical approach to this technoscientific field. On the one hand, ethical issues have been identified, quantitatively analyzed and categorized in three main thematic groups. Namely: Well-being, Care, and Justice. On the other hand-and on the basis of some significant disclosed tendencies of the current approach-, future lines of research and issues regarding the enrichment of the ethical gaze on social assistive robotics have been identified and outlined.
Assistive robots need to be able to perform a large number of tasks that imply some type of cloth... more Assistive robots need to be able to perform a large number of tasks that imply some type of cloth manipulation. These tasks include domestic chores such as laundry handling or bed-making, among others, as well as dressing assistance to disabled users. Due to the deformable nature of fabrics, this manipulation requires a strong perceptual feedback. Common perceptual skills that enable robots to complete their cloth manipulation tasks are reviewed here, mainly relying on vision, but also resorting to touch and force. The use of such basic skills is then examined in the context of the different cloth manipulation tasks, be them garment-only applications in the line of performing domestic chores, or involving physical contact with a human as in dressing assistance. Keywords Robotic assistance • Robotic cloth manipulation • Perception of cloth 1 Introduction Robots perform quite competently nowadays in structured environments, tackling hard tasks under tough working conditions, and even handling incidences that could be anticipated. The requirements posed by assistive settings, however, point in a quite different direction. The top concerns are no longer precision and repeatability, but rather a high degree of adaptability to varying ambient conditions, ability to learn, multimodal human-robot interaction capabilities, and integrated safety. It is obvious that robots cannot replace humans entirely in assistive environments, probably they shouldn't either. Nonetheless, it is desirable that they be able to perform a variety of tasks within domestic and service environments such as hospitals or care homes. Such tasks fall under what is commonly known as domestic chores, whose fulfillment ensures not just tidy homes, but the very proper life conditions of physically disabled people, in acceptable standards of human dignity and coverage of needs. Human companionship cannot be obviated, but the burden of associated duties without added value in personal interchange can be certainly alleviated with a robotic helper. Among these tasks, those that involve the manipulation of textile items, including of course all types of garments, but also other categories of fabric-made objects such as bed-and tablecloth, curtains, towels, kitchen rags and dishcloth, etc. have to be highlighted. The omnipresence of such items in human daily environments and the importance of handling them correctly are evident. The means of providing assistive robots with the necessary abilities to perform cloth manipulation are not so clear. Ideally one would like to replicate the human proficiency in manipulating clothes, but the robotic state-of-the-art is still far from achieving the needed perceptual skills and the required dexterity. Cloth perception and manipulation
It is argued in this paper that motion planning and control inyolye two different types of proces... more It is argued in this paper that motion planning and control inyolye two different types of processing. Global planning such as that required to make one's way through a city inyolyes symbolic sequential processing, whilst local obstacle-ayoidance capabilities such as those used to reach one's mug when there are other objects around inYolYe massiYely-parallel distributed processing. The main algorithms for motion planning and control deyeloped within the AI and neurocomputing paradigms are reviewed, their strengths and weaknesses being highlighted. The conclusion is that a combination of AI techniques with neurocomputing ones that takes advantage of the strong points of both computing paradigms is required to deyelop robust computational systems for motion planning and control. The key point to achieve such a combination is to devise good interfaces between the representations used in the two paradigms.
Robots that can assist in the activities of daily living, such as dressing, may support older adu... more Robots that can assist in the activities of daily living, such as dressing, may support older adults, addressing the needs of an aging population in the face of a growing shortage of care professionals. Using depth cameras during robot-assisted dressing can lead to occlusions and loss of user tracking, which may result in unsafe trajectory planning or prevent the planning task proceeding altogether. For the dressing task of putting on a jacket, which is addressed in this letter, tracking of the arm is lost when the user's hand enters the jacket, which may lead to unsafe situations for the user and a poor interaction experience. Using motion tracking data, free from occlusions, gathered from a human-human interaction study on an assisted dressing task, recurrent neural network models were built to predict the elbow position of a single arm based on other features of the user pose. The best features for predicting the elbow position were explored by using regression trees indicating the hips and shoulder as possible predictors. Engineered features were also created based on observations of real dressing scenarios and their effectiveness explored. Comparison between position and orientation-based datasets was also included in this study. A 12-fold cross-validation was performed for each feature set and repeated 20 times to improve statistical power. Using position-based data, the elbow position could be predicted with a 4.1 cm error but adding engineered features reduced the error to 2.4 cm. Adding orientation information to the data did not improve the accuracy and aggregating univariate response models failed to make significant improvements. The model was evaluated on Kinect data for a robot dressing task and although not without issues, demonstrates potential for this application. Although this has been demonstrated for jacket dressing, the technique could be applied to a number of different situations during occluded tracking. Index Terms-Physical human-robot interaction, human detection and tracking, deep learning in robotics and automation, robot safety.
Movement Primitives (MPs) have been widely adopted for representing and learning robotic movement... more Movement Primitives (MPs) have been widely adopted for representing and learning robotic movements using Reinforcement Learning Policy Search. Probabilistic Movement Primitives (ProMPs) are a kind of MP based on a stochastic representation over sets of trajectories, able of capturing the variability allowed while executing a movement. This approach has proved effective in learning a wide range of robotic movements, but it comes with the need of dealing with a high-dimensional space of parameters. This may be a critical problem when learning tasks with two robotic manipulators, and this work proposes an approach to reduce the dimension of the parameter space based on the exploitation of symmetry. A symmetrization method for ProMPs is presented and used to represent two movements, employing a single ProMP for the first arm and a symmetry surface that maps that ProMP to the second arm. This symmetric representation is then adopted in reinforcement learning of bimanual tasks (from user-provided demonstrations), using Relative Entropy Policy Search (REPS) algorithm. The symmetry-based approach developed has been tested in an experiment of cloth manipulation, showing a speed increment in learning the task.
This article presents a new method for actively exploring a 3D workspace with the aim of localizi... more This article presents a new method for actively exploring a 3D workspace with the aim of localizing relevant regions for a given task. Our method encodes the exploration route in a multi-layer occupancy grid map. This map, together with a multiple-view estimator and a maximum-information-gain gathering approach, incrementally provide a better understanding of the scene until reaching the task termination criterion. This approach is designed to be applicable to any task entailing 3D object exploration where some previous knowledge of its approximate shape is available. Its suitability is demonstrated here for a leaf probing task using an eye-in-hand arm configuration in the context of a phenotyping application (leaf probing).
Series in Machine Perception and Artificial Intelligence
... A change of configuration is here expressed through a rotation of angle£ around a rotation ax... more ... A change of configuration is here expressed through a rotation of angle£ around a rotation axis given by the unit vector k=(kx, kv, kx) and a translation t=(tx, tv, t,). To attain a rotation of angle£/, a path is taken which is not exactly linear in rotational displacements but constitutes a ...
Deformable object (e.g., clothes) manipulation by a robot in interaction with a human being prese... more Deformable object (e.g., clothes) manipulation by a robot in interaction with a human being presents several interesting challenges. Due to texture and deformability, the object can get hooked in the human limbs. Moreover, the human can change their limbs position and curvature, which require changes in the paths to be followed by the robot. To help solve these problems, in this paper we propose a technique of learning by demonstration able to adapt to changes in position and curvature of the object (human limb) and recover from execution faults (hooks). The technique is tested using simulations, but with data obtained from a real robot. 1
In general, rearranging the legs of a Stewart-Gough platform, i.e., changing the locations of its... more In general, rearranging the legs of a Stewart-Gough platform, i.e., changing the locations of its leg attachments, modifies the platform singularity locus in a rather unexpected way. Nevertheless, some leg rearrangements have been recently found to leave singularities invariant but, unfortunately, these rearrangements are only valid for Stewart-Gough platforms containing rigid components. In this work, the authors go a step further presenting singularity-invariant leg rearrangements that can be applied to any Stewart-Gough platform whose base and platform attachments are coplanar. The practical consequences of the presented theoretical results are illustrated with several examples including well-known architectures.
We consider the following problem: given any two convex polygons which are free to translate and ... more We consider the following problem: given any two convex polygons which are free to translate and rotate arbitrarily in the plane and which have m and n vertices respectively, what is the maximum number of ways they may contact each other such that three independent boundary contacts are made? This problem is similar to several others which have been investigated in the eld of combinatorial geometry, such as the polygon containment problem, but has the signiicant diierence that in our context the polygons are allowed to overlap. We present the results of an empirical study of the number of such conngurations, and we prove that an upper bound on the total is O(m 3 n 2 + m 2 n 3).
Proceedings of the Fourth International Conference on Computer Vision Theory and Applications
We register close-range depth images of objects using a Swissranger sensor and apply a spring-mas... more We register close-range depth images of objects using a Swissranger sensor and apply a spring-mass model for 3D object reconstruction. The Swissranger sensor delivers depth images in real time which have, compared with other types of sensors, such as laser scanners, a lower resolution and are afflicted with larger uncertainties. To reduce noise and remove outliers in the data, we treat the point cloud as a system of interacting masses connected via elastic forces. We investigate two models, one with and one without a surface-topology preserving interaction strength. The algorithm is applied to synthetic and real Swissranger sensor data, demonstrating the feasibility of the approach. This method represents a preliminary step before fitting higher-level surface descriptors to the data, which will be required to define object-action complexes (OACS) for robot applications.
The ability to deal with articulated objects is very important for robots assisting humans. In th... more The ability to deal with articulated objects is very important for robots assisting humans. In this work, a framework to robustly and adaptively operate common doors, using an autonomous mobile manipulator, is proposed. To push forward the state-of-the-art in robustness and speed performance, we devise a novel algorithm that fuses a convolutional neural network with efficient point cloud processing. This advancement enables real-time grasping pose estimation for multiple handles from RGB-D images, providing a speed up improvement for assistive human-centered applications. In addition, we propose a versatile Bayesian framework that endows the robot with the ability to infer the door kinematic model from observations of its motion and learn from previous experiences or human demonstrations. Combining these algorithms with a Task Space Region motion planner, we achieve efficient door operation regardless of the kinematic model. We validate our framework with real-world experiments usin...
The expectation of a function of random variables is modelled as the value of the function in the... more The expectation of a function of random variables is modelled as the value of the function in the mean value of the variables plus a penalty term. This penalty term is calculated exactly, and the properties of different approximations are analyzed. In particular, for quadratic functions, the penalty term is shown to have a supervised and an unsupervised part. Then, two algorithms for minimizing the expected error of a feedforward network of random weights are compared. One of them is stochastical, while the other is deterministic and more flexible. Given a particular feedforward network architecture and a training set, the deterministic algorithm here presented accurately finds the weight configuration that makes the network response most resistant to a class of weight perturbations. Finally, the study of the most stable configurations of a network unravels some undesirable properties of networks with asymmetric activation functions
Es presenta un metode per estimar el moviment d’un robot a partir del seguiment de contorns actiu... more Es presenta un metode per estimar el moviment d’un robot a partir del seguiment de contorns actius en una successio d’imatges adquirides amb un sistema monocular i calibrat, muntat sobre el propi robot. Un cop el contorn actiu ha estat ajustat al voltant de la projeccio d’un objecte al pla de la imatge, les deformacions afins que pateix el contorn en la sequencia es parametritzen utilitzant un vector de forma de dimensio constant. S’analitzen les implicacions d’utilitzar vectors de forma de dimensio 3, 4 i 6. A partir d’aquests vectors de forma es pot trobar la rotacio i translacio 3D que ha provocat les deformacions, es a dir, el moviment propi de la camera. Els resultats preliminars obtinguts amb un robot rodant sobre un terra pla i vectors de forma de 4 components son forca encoratjadors.
In this work we propose a decision-making system that efficiently learns behaviors in the form of... more In this work we propose a decision-making system that efficiently learns behaviors in the form of rules using natural human instructions about cause-effect relations in currently observed situations, avoiding complicated instructions and explanations of long-run action sequences and complete world dynamics. The learned rules are represented in a way suitable to both reactive and deliberative approaches, which are thus smoothly integrated. Simple and repetitive tasks are resolved reactively, while complex tasks would be faced in a more deliberative manner using a planner module. Human interaction is only required if the system fails to obtain the expected results when applying a rule, or fails to resolve the task with the knowledge acquired so far. I. INTRODUCTION N this work we are facing the problem of decision making for a multitask robot embedded in a human environment that should rapidly learn to perform tasks by interacting with humans, in an on-line way, and without any previo...
This work presents an evaluation of three color constancy techniques applied to a landmark detect... more This work presents an evaluation of three color constancy techniques applied to a landmark detection system designed for a walking robot, which has to operate in unknown and unstructured outdoor environments. The first technique is the well-known image conversion to a chromaticity space, and the second technique is based on successive lighting intensity and illuminant color normalizations. Based on a differential model of color constancy, we propose the third technique, based on color ratios, which unifies the processes of color constancy and landmark detection. The approach used to detect potential landmarks, which is common to all evaluated systems, is based on visual saliency concepts using multiscale color opponent features to identify salient regions in the images. These regions are selected as landmark candidates, and they are further characterized by their features for identification and recognition.
Resumen: La cuestion de la decidibilidad, es decir, si es posible demostrar computacionalmente qu... more Resumen: La cuestion de la decidibilidad, es decir, si es posible demostrar computacionalmente que una expresion matematica es verdadera o falsa, fue planteada por Hilbert y permanecio abierta hasta que Turing la respondio de forma negativa. Establecida la no-decidibilidad de las matematicas, los esfuerzos en informatica teorica se centraron en el estudio de la complejidad computacional de los problemas decidibles. En este articulo presentamos una breve introduccion a las clases P (problemas resolubles en tiempo polinomico) y nP (problemas resolubles de manera no determinista en tiempo polinomico), al tiempo que exponemos la dificultad de establecer si P = nP y las consecuencias que se derivarian de que ambas clases de problemas fueran iguales. Esta cuestion tiene implicaciones no solo en los campos de la informatica, las matematicas y la fisica, sino tambien para la biologia, la sociologia y la economia. La idea seminal del estudio de la complejidad computacional es consecuencia di...
Along with its potential contributions to the practice of care, social assistive robotics raises ... more Along with its potential contributions to the practice of care, social assistive robotics raises significant ethical issues. The growing development of this technoscientific field of intelligent robotics has thus triggered a widespread proliferation of ethical attention towards its disruptive potential. However, the current landscape of ethical debate is fragmented and conceptually disordered, endangering ethics' practical strength for normatively addressing these challenges. This paper presents a critical literature review of the ethical issues of social assistive robotics, which provides a comprehensive and intelligible overview of the current ethical approach to this technoscientific field. On the one hand, ethical issues have been identified, quantitatively analyzed and categorized in three main thematic groups. Namely: Well-being, Care, and Justice. On the other hand-and on the basis of some significant disclosed tendencies of the current approach-, future lines of research and issues regarding the enrichment of the ethical gaze on social assistive robotics have been identified and outlined.
Assistive robots need to be able to perform a large number of tasks that imply some type of cloth... more Assistive robots need to be able to perform a large number of tasks that imply some type of cloth manipulation. These tasks include domestic chores such as laundry handling or bed-making, among others, as well as dressing assistance to disabled users. Due to the deformable nature of fabrics, this manipulation requires a strong perceptual feedback. Common perceptual skills that enable robots to complete their cloth manipulation tasks are reviewed here, mainly relying on vision, but also resorting to touch and force. The use of such basic skills is then examined in the context of the different cloth manipulation tasks, be them garment-only applications in the line of performing domestic chores, or involving physical contact with a human as in dressing assistance. Keywords Robotic assistance • Robotic cloth manipulation • Perception of cloth 1 Introduction Robots perform quite competently nowadays in structured environments, tackling hard tasks under tough working conditions, and even handling incidences that could be anticipated. The requirements posed by assistive settings, however, point in a quite different direction. The top concerns are no longer precision and repeatability, but rather a high degree of adaptability to varying ambient conditions, ability to learn, multimodal human-robot interaction capabilities, and integrated safety. It is obvious that robots cannot replace humans entirely in assistive environments, probably they shouldn't either. Nonetheless, it is desirable that they be able to perform a variety of tasks within domestic and service environments such as hospitals or care homes. Such tasks fall under what is commonly known as domestic chores, whose fulfillment ensures not just tidy homes, but the very proper life conditions of physically disabled people, in acceptable standards of human dignity and coverage of needs. Human companionship cannot be obviated, but the burden of associated duties without added value in personal interchange can be certainly alleviated with a robotic helper. Among these tasks, those that involve the manipulation of textile items, including of course all types of garments, but also other categories of fabric-made objects such as bed-and tablecloth, curtains, towels, kitchen rags and dishcloth, etc. have to be highlighted. The omnipresence of such items in human daily environments and the importance of handling them correctly are evident. The means of providing assistive robots with the necessary abilities to perform cloth manipulation are not so clear. Ideally one would like to replicate the human proficiency in manipulating clothes, but the robotic state-of-the-art is still far from achieving the needed perceptual skills and the required dexterity. Cloth perception and manipulation
It is argued in this paper that motion planning and control inyolye two different types of proces... more It is argued in this paper that motion planning and control inyolye two different types of processing. Global planning such as that required to make one's way through a city inyolyes symbolic sequential processing, whilst local obstacle-ayoidance capabilities such as those used to reach one's mug when there are other objects around inYolYe massiYely-parallel distributed processing. The main algorithms for motion planning and control deyeloped within the AI and neurocomputing paradigms are reviewed, their strengths and weaknesses being highlighted. The conclusion is that a combination of AI techniques with neurocomputing ones that takes advantage of the strong points of both computing paradigms is required to deyelop robust computational systems for motion planning and control. The key point to achieve such a combination is to devise good interfaces between the representations used in the two paradigms.
Robots that can assist in the activities of daily living, such as dressing, may support older adu... more Robots that can assist in the activities of daily living, such as dressing, may support older adults, addressing the needs of an aging population in the face of a growing shortage of care professionals. Using depth cameras during robot-assisted dressing can lead to occlusions and loss of user tracking, which may result in unsafe trajectory planning or prevent the planning task proceeding altogether. For the dressing task of putting on a jacket, which is addressed in this letter, tracking of the arm is lost when the user's hand enters the jacket, which may lead to unsafe situations for the user and a poor interaction experience. Using motion tracking data, free from occlusions, gathered from a human-human interaction study on an assisted dressing task, recurrent neural network models were built to predict the elbow position of a single arm based on other features of the user pose. The best features for predicting the elbow position were explored by using regression trees indicating the hips and shoulder as possible predictors. Engineered features were also created based on observations of real dressing scenarios and their effectiveness explored. Comparison between position and orientation-based datasets was also included in this study. A 12-fold cross-validation was performed for each feature set and repeated 20 times to improve statistical power. Using position-based data, the elbow position could be predicted with a 4.1 cm error but adding engineered features reduced the error to 2.4 cm. Adding orientation information to the data did not improve the accuracy and aggregating univariate response models failed to make significant improvements. The model was evaluated on Kinect data for a robot dressing task and although not without issues, demonstrates potential for this application. Although this has been demonstrated for jacket dressing, the technique could be applied to a number of different situations during occluded tracking. Index Terms-Physical human-robot interaction, human detection and tracking, deep learning in robotics and automation, robot safety.
Movement Primitives (MPs) have been widely adopted for representing and learning robotic movement... more Movement Primitives (MPs) have been widely adopted for representing and learning robotic movements using Reinforcement Learning Policy Search. Probabilistic Movement Primitives (ProMPs) are a kind of MP based on a stochastic representation over sets of trajectories, able of capturing the variability allowed while executing a movement. This approach has proved effective in learning a wide range of robotic movements, but it comes with the need of dealing with a high-dimensional space of parameters. This may be a critical problem when learning tasks with two robotic manipulators, and this work proposes an approach to reduce the dimension of the parameter space based on the exploitation of symmetry. A symmetrization method for ProMPs is presented and used to represent two movements, employing a single ProMP for the first arm and a symmetry surface that maps that ProMP to the second arm. This symmetric representation is then adopted in reinforcement learning of bimanual tasks (from user-provided demonstrations), using Relative Entropy Policy Search (REPS) algorithm. The symmetry-based approach developed has been tested in an experiment of cloth manipulation, showing a speed increment in learning the task.
This article presents a new method for actively exploring a 3D workspace with the aim of localizi... more This article presents a new method for actively exploring a 3D workspace with the aim of localizing relevant regions for a given task. Our method encodes the exploration route in a multi-layer occupancy grid map. This map, together with a multiple-view estimator and a maximum-information-gain gathering approach, incrementally provide a better understanding of the scene until reaching the task termination criterion. This approach is designed to be applicable to any task entailing 3D object exploration where some previous knowledge of its approximate shape is available. Its suitability is demonstrated here for a leaf probing task using an eye-in-hand arm configuration in the context of a phenotyping application (leaf probing).
Series in Machine Perception and Artificial Intelligence
... A change of configuration is here expressed through a rotation of angle£ around a rotation ax... more ... A change of configuration is here expressed through a rotation of angle£ around a rotation axis given by the unit vector k=(kx, kv, kx) and a translation t=(tx, tv, t,). To attain a rotation of angle£/, a path is taken which is not exactly linear in rotational displacements but constitutes a ...
Deformable object (e.g., clothes) manipulation by a robot in interaction with a human being prese... more Deformable object (e.g., clothes) manipulation by a robot in interaction with a human being presents several interesting challenges. Due to texture and deformability, the object can get hooked in the human limbs. Moreover, the human can change their limbs position and curvature, which require changes in the paths to be followed by the robot. To help solve these problems, in this paper we propose a technique of learning by demonstration able to adapt to changes in position and curvature of the object (human limb) and recover from execution faults (hooks). The technique is tested using simulations, but with data obtained from a real robot. 1
In general, rearranging the legs of a Stewart-Gough platform, i.e., changing the locations of its... more In general, rearranging the legs of a Stewart-Gough platform, i.e., changing the locations of its leg attachments, modifies the platform singularity locus in a rather unexpected way. Nevertheless, some leg rearrangements have been recently found to leave singularities invariant but, unfortunately, these rearrangements are only valid for Stewart-Gough platforms containing rigid components. In this work, the authors go a step further presenting singularity-invariant leg rearrangements that can be applied to any Stewart-Gough platform whose base and platform attachments are coplanar. The practical consequences of the presented theoretical results are illustrated with several examples including well-known architectures.
We consider the following problem: given any two convex polygons which are free to translate and ... more We consider the following problem: given any two convex polygons which are free to translate and rotate arbitrarily in the plane and which have m and n vertices respectively, what is the maximum number of ways they may contact each other such that three independent boundary contacts are made? This problem is similar to several others which have been investigated in the eld of combinatorial geometry, such as the polygon containment problem, but has the signiicant diierence that in our context the polygons are allowed to overlap. We present the results of an empirical study of the number of such conngurations, and we prove that an upper bound on the total is O(m 3 n 2 + m 2 n 3).
Uploads
Papers by Carme Torras