Manschitz, Simon (2018)
Learning Sequential Skills for Robot Manipulation Tasks.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
Most people's imagination about robots has been shaped by Hollywood movies or novels, resulting in the dream of having robots as assistants or household helpers in our homes. However, there is still a large gap between this dream and the actual capabilities of robots. One underlying reason is that every home is unique and largely unstructured, making it impossible to pre-program a robot for all the challenges it might face in such an environment. For instance, floor plans and furniture differ from home to home. Humans and pets walk around, potentially getting in the robot's way and making the environment non-static. Hence, a pre-programmed robot deployed in such an environment will undoubtedly face problems that it cannot solve with its existing knowledge. In order to cope with this issue, researchers started to equip robots with learning capabilities. Ideally, such capabilities allow a robot to adapt skills to new or changing situations or even to learn completely new tasks. Also humans learn new skills over time and are able to adapt them if needed. Therefore, such learning capabilities seem natural to us. If we are not able to master a specific task, we usually would ask another person to demonstrate it or to give instructions on how to perform it. In robotics research, the field of "Learning from Demonstration" tries to mimic this behavior by learning new skills from demonstrations of a task. By applying machine learning techniques, the data perceived from a single or multiple demonstrations are exploited to learn a mapping from perception to the action of a robot.
In this thesis, we concentrate on important Learning from Demonstration aspects that have not gotten so much attention in the research community so far. In particular, we focus on learning methods for robot manipulation tasks. These tasks have two important characteristics. First, they can be naturally decomposed into a set of subtasks and, therefore, can be mastered by performing the individual subtasks in the correct sequential order. Second, they involve physical contact between the robot and objects in its environment. One aim of this thesis is developing methods which allow for learning skills for robot manipulation tasks that generalize well to unknown situations. For instance, a learned skill should also be applicable if positions and orientations of objects differ from those seen in a demonstration.
In the first part of the thesis, we focus on the "sequential" aspect of manipulation tasks. Many approaches assume that subtasks are executed in a purely sequential manner or that the human always demonstrates the same sequence of subtasks. We propose an approach that does not have this assumption. Based on the demonstrations, a graph is generated which connects the subtasks with each other. Each subtask is associated with a movement primitive, a basic elementary movement necessary to perform the subtask. Depending on the environmental conditions, different sequences of movement primitives are executed, allowing the robot to perform tasks which for instance require an arbitrary number of repetitions (e.g., unscrewing a light bulb).
As we concentrate on the sequential aspects of a task in the first part of the thesis, we assume the demonstrations are labeled with the correct movement primitives over time. Additionally, the movement primitives are predefined. In the second part of the thesis, these two assumptions are relaxed. We first present an approach which decomposes the demonstrations into a set of meaningful movement primitives by inferring the underlying sequential structure of the task. The decomposition is based on a probability distribution we call Directional Normal Distribution. By utilizing the distribution, our method infers if a movement should be performed relative to an object in the scene and if a force should be applied in certain directions or not. Forces are especially important when interacting with the environment, for example if the robot has to manipulate objects. By defining movements relative to objects in the scene, the robot is likely to generalize better to new situations, for instance if the object positions differ from the demonstrations. Our task-decomposition method allows for inferring the most likely movement primitives over time and replaces the process of manually labeling the demonstrations. By combining the method with the sequencing concept presented in the first part of the thesis, complex skills can be learned from scratch without further human supervision. Such a learning scheme is an essential requirement for domestic robots, as not every human teacher might be able or willing to do the tedious labeling of the data.
In both the decomposition and the sequencing part of the thesis, we assume that the teacher performs point-to-point movements and stops between two successive movements. While these assumptions lead to an approach which can learn skills for fairly complex tasks, it also restricts the class of tasks for which the approach can be used. In the third part of the thesis, we therefore introduce the Mixture of Attractors movement primitive representation. Here, a movement is modulated by continuously changing the activations of a set of simple attractors over time. We present a learning algorithm for the representation which learns both the attractors and their activations. An important property of the representation is that the attractors can be defined in different coordinate frames. The continuous activations and the attractors defined in different coordinate frames allow the system to learn movements of arbitrary shape and to generalize them to different object positions. In addition, the transitions between successive movements are smooth. This property reflects an important behavior of humans who often tend to co-articulate between successive movements. In contrast to many existing approaches, movements are learned by solving a convex optimization problem that does not rely on a good initial estimate of parameters.
In summary, the contribution of this thesis to the state-of-the-art in Learning from Demonstration is two-fold. The first contribution is a framework which is able to learn sequential skills for robot manipulation tasks from a few demonstrations. In contrast to other approaches, our method incorporates object-relative movements and force information directly into the skill learning framework. The second contribution is the Mixture of Attractors movement primitive representation. The representation supports co-articulated movements represented in different coordinate frames and outperforms existing movement primitive representations in terms of accuracy and generalization capabilities. Both contributions are evaluated on a wide range of tasks in simulation and on a real single arm robot with seven degrees of freedom. Altogether, this thesis aims at bringing us closer to the dream of having autonomous robots in our homes.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2018 | ||||
Autor(en): | Manschitz, Simon | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Learning Sequential Skills for Robot Manipulation Tasks | ||||
Sprache: | Englisch | ||||
Referenten: | Peters, Prof. Dr. Jan ; Billard, Prof. Dr. Aude ; Kober, Dr.-Ing. Jens | ||||
Publikationsjahr: | 2018 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 19 Dezember 2017 | ||||
URL / URN: | http://tuprints.ulb.tu-darmstadt.de/7185 | ||||
Kurzbeschreibung (Abstract): | Most people's imagination about robots has been shaped by Hollywood movies or novels, resulting in the dream of having robots as assistants or household helpers in our homes. However, there is still a large gap between this dream and the actual capabilities of robots. One underlying reason is that every home is unique and largely unstructured, making it impossible to pre-program a robot for all the challenges it might face in such an environment. For instance, floor plans and furniture differ from home to home. Humans and pets walk around, potentially getting in the robot's way and making the environment non-static. Hence, a pre-programmed robot deployed in such an environment will undoubtedly face problems that it cannot solve with its existing knowledge. In order to cope with this issue, researchers started to equip robots with learning capabilities. Ideally, such capabilities allow a robot to adapt skills to new or changing situations or even to learn completely new tasks. Also humans learn new skills over time and are able to adapt them if needed. Therefore, such learning capabilities seem natural to us. If we are not able to master a specific task, we usually would ask another person to demonstrate it or to give instructions on how to perform it. In robotics research, the field of "Learning from Demonstration" tries to mimic this behavior by learning new skills from demonstrations of a task. By applying machine learning techniques, the data perceived from a single or multiple demonstrations are exploited to learn a mapping from perception to the action of a robot. In this thesis, we concentrate on important Learning from Demonstration aspects that have not gotten so much attention in the research community so far. In particular, we focus on learning methods for robot manipulation tasks. These tasks have two important characteristics. First, they can be naturally decomposed into a set of subtasks and, therefore, can be mastered by performing the individual subtasks in the correct sequential order. Second, they involve physical contact between the robot and objects in its environment. One aim of this thesis is developing methods which allow for learning skills for robot manipulation tasks that generalize well to unknown situations. For instance, a learned skill should also be applicable if positions and orientations of objects differ from those seen in a demonstration. In the first part of the thesis, we focus on the "sequential" aspect of manipulation tasks. Many approaches assume that subtasks are executed in a purely sequential manner or that the human always demonstrates the same sequence of subtasks. We propose an approach that does not have this assumption. Based on the demonstrations, a graph is generated which connects the subtasks with each other. Each subtask is associated with a movement primitive, a basic elementary movement necessary to perform the subtask. Depending on the environmental conditions, different sequences of movement primitives are executed, allowing the robot to perform tasks which for instance require an arbitrary number of repetitions (e.g., unscrewing a light bulb). As we concentrate on the sequential aspects of a task in the first part of the thesis, we assume the demonstrations are labeled with the correct movement primitives over time. Additionally, the movement primitives are predefined. In the second part of the thesis, these two assumptions are relaxed. We first present an approach which decomposes the demonstrations into a set of meaningful movement primitives by inferring the underlying sequential structure of the task. The decomposition is based on a probability distribution we call Directional Normal Distribution. By utilizing the distribution, our method infers if a movement should be performed relative to an object in the scene and if a force should be applied in certain directions or not. Forces are especially important when interacting with the environment, for example if the robot has to manipulate objects. By defining movements relative to objects in the scene, the robot is likely to generalize better to new situations, for instance if the object positions differ from the demonstrations. Our task-decomposition method allows for inferring the most likely movement primitives over time and replaces the process of manually labeling the demonstrations. By combining the method with the sequencing concept presented in the first part of the thesis, complex skills can be learned from scratch without further human supervision. Such a learning scheme is an essential requirement for domestic robots, as not every human teacher might be able or willing to do the tedious labeling of the data. In both the decomposition and the sequencing part of the thesis, we assume that the teacher performs point-to-point movements and stops between two successive movements. While these assumptions lead to an approach which can learn skills for fairly complex tasks, it also restricts the class of tasks for which the approach can be used. In the third part of the thesis, we therefore introduce the Mixture of Attractors movement primitive representation. Here, a movement is modulated by continuously changing the activations of a set of simple attractors over time. We present a learning algorithm for the representation which learns both the attractors and their activations. An important property of the representation is that the attractors can be defined in different coordinate frames. The continuous activations and the attractors defined in different coordinate frames allow the system to learn movements of arbitrary shape and to generalize them to different object positions. In addition, the transitions between successive movements are smooth. This property reflects an important behavior of humans who often tend to co-articulate between successive movements. In contrast to many existing approaches, movements are learned by solving a convex optimization problem that does not rely on a good initial estimate of parameters. In summary, the contribution of this thesis to the state-of-the-art in Learning from Demonstration is two-fold. The first contribution is a framework which is able to learn sequential skills for robot manipulation tasks from a few demonstrations. In contrast to other approaches, our method incorporates object-relative movements and force information directly into the skill learning framework. The second contribution is the Mixture of Attractors movement primitive representation. The representation supports co-articulated movements represented in different coordinate frames and outperforms existing movement primitive representations in terms of accuracy and generalization capabilities. Both contributions are evaluated on a wide range of tasks in simulation and on a real single arm robot with seven degrees of freedom. Altogether, this thesis aims at bringing us closer to the dream of having autonomous robots in our homes. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-71850 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik | ||||
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik > Intelligente Autonome Systeme 20 Fachbereich Informatik |
||||
Hinterlegungsdatum: | 04 Feb 2018 20:55 | ||||
Letzte Änderung: | 04 Feb 2018 20:55 | ||||
PPN: | |||||
Referenten: | Peters, Prof. Dr. Jan ; Billard, Prof. Dr. Aude ; Kober, Dr.-Ing. Jens | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 19 Dezember 2017 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |