Kirchhoff, Jérôme (2017)
Towards Dependability of Ultra Lightweight Tendon Driven Series Elastic Robots.
Technische Universität Darmstadt
Dissertation, Erstveröffentlichung
Kurzbeschreibung (Abstract)
Ultra lightweight tendon driven series elastic robots exhibit unique high safety features based on very low moving link masses and mechanical decoupling of actuator inertias from links. These make them highly suitable for safe physical human-robot interaction and collaboration. However, such actuation approach introduces a number of challenges to be overcome in order to capitalize on their advantages. These include some inherent uncertainties regarding their dependability, which mainly result from the combination of possible changes in the drive train caused by wear, unavoidable system model inaccuracies, and the high mechanical elasticities facilitating oscillations in the joints. It is shown that the resulting challenges, in particular influencing the performance in velocity estimation, robust torque transmission, trajectory control, and human-robot collaboration, can to a large extent be overcome. Thereby, achieving dependable usage of such robots marks a key step towards utilizing their unique high safety features for physical human-robot collaboration in industrial applications. In order to get an insight into the safety capabilities as a collaborative robot, a risk analysis according to the new ISO/TS 15066 has been performed comparing the class of tendon driven ultra lightweight series elastic robots with downscaled and stiff robots. Further, the influence of high compliance during contact situations has been demonstrated during a collision test.
In lightweight and downscaled robotic structures, the suitable joint position sensor size is limited, resulting in a relatively coarse discrete position signal. Furthermore, joint elasticities facilitate fast and oscillating motions containing a broad bandwidth of frequencies and velocities. In order to effectively damp the controlled motion and also for model-based computations, an accurate velocity estimation is essential. Kalman filter approaches already showed accurate performance in position signal based velocity estimation, but within a small bandwidth that correlates with the filter's measurement variance parameter. In order to create an estimation approach more suitable for the considered concept of tendon driven series elastic actuation, the velocity and frequency dependent optimal measurement variance parameter has been analyzed. The observed parameter characteristics serve as the basis for the proposed new measurement variance update rule. The resulting novel adaptive Kalman filter approach adapts better to the investigated application than the compared state-of-the-art approach. It produces a smoother and more accurate velocity estimation, as will be demonstrated in virtual and real robot experiments.
In biologically inspired mechanical structures, tendons are used to transmit forces along a kinematic chain, which must satisfy high robustness requirements by simultaneously allowing small pulley radii. Cables or belts are often not applicable due to size and force requirements in contrast to thin synthetic fiber ropes. Besides the breaking force, only a little information about the ropes is provided by the manufacturer. Thus, further research is required to investigate whether a rope is suitable to be used as a reliable component of a drive train which mainly depends on its elongation behavior. In this regard, new systematic creep experiments regarding different materials, manufacturers, and diameters, as well as bending experiments are presented in this work. The findings obtained for the rope characteristics support the material selection decisions during the system design process and give insights into the long-term behavior. In order to monitor the long-term behavior, an observer approach is presented which does not need joint torque measurements but nevertheless enables elongation detection in tendon driven kinematic chains even subject to model inaccuracies. This has been demonstrated in simulated and real robot experiments.
In an industrial application, a robotic system is expected to perform a motion without alterations over time accurately. The inherent uncertainties of the elastic tendon driven actuation can strongly influence the control performance and, thus, the reliability of the robot motion. State-of-the-art control approaches for joint elastic robot arms rely on accurate joint torque measurements and accurate models of drive train dynamics. Available torque sensors with a suitable torque range are heavy and large and do not fit into the lightweight actuation. Thus, so far only erroneous model-based joint torque estimation and consequently inaccurate load dependent friction models can be used, for which these control approaches perform unsatisfactorily. With the aim to reduce the model dependency, the performance of a friction observer under joint torque estimation errors is investigated in this thesis. This leads to the final proposed controller design that is able to compensate for both, drive train and load dynamics model inaccuracies. In particular, it integrates a friction observer into a state space approach without the need for explicit joint torque measurements eliminating the steady-state control errors and realizes contact handling. The resulting control performance has been evaluated in simulated and real robot experiments demonstrating an accurate and damped motion performance.
An essential part of human-robot collaboration consists of physical interaction for joint task solving. To perceive physical interactions, the robotic arm must be able to estimate external forces. This can be achieved by disturbance observers, which only use internal data and are therefore suitable for the regarded system. In this thesis, the accuracy limitations are investigated and discussed theoretically and via simulated robot experiments to reveal the capabilities of tendon driven joint elastic robots regarding external torque estimations. Furthermore, the user should be able to teach the needed robot motion for the executed process without high effort. For this purpose, a teach-in controller that enables hand guided motions even under drive train model errors is presented and evaluated in realistic simulation. Besides the physical interaction, successful collaboration requires clear communication between the human and robot. To investigate the influence of robot state visualizations on the user's situational awareness and robot’s intention estimation in collaborative task solving, a preliminary user study is conducted. Further, metrics for wear estimation are introduced and discussed to support the human during the assessment of the robot's integrity and serve to increase trust in collaboration.
Typ des Eintrags: | Dissertation | ||||
---|---|---|---|---|---|
Erschienen: | 2017 | ||||
Autor(en): | Kirchhoff, Jérôme | ||||
Art des Eintrags: | Erstveröffentlichung | ||||
Titel: | Towards Dependability of Ultra Lightweight Tendon Driven Series Elastic Robots | ||||
Sprache: | Englisch | ||||
Referenten: | von Stryk, Prof. Dr. Oskar ; Wollherr, PD Dr. Dirk | ||||
Publikationsjahr: | Januar 2017 | ||||
Ort: | Darmstadt | ||||
Datum der mündlichen Prüfung: | 16 August 2017 | ||||
URL / URN: | http://tuprints.ulb.tu-darmstadt.de/7637 | ||||
Kurzbeschreibung (Abstract): | Ultra lightweight tendon driven series elastic robots exhibit unique high safety features based on very low moving link masses and mechanical decoupling of actuator inertias from links. These make them highly suitable for safe physical human-robot interaction and collaboration. However, such actuation approach introduces a number of challenges to be overcome in order to capitalize on their advantages. These include some inherent uncertainties regarding their dependability, which mainly result from the combination of possible changes in the drive train caused by wear, unavoidable system model inaccuracies, and the high mechanical elasticities facilitating oscillations in the joints. It is shown that the resulting challenges, in particular influencing the performance in velocity estimation, robust torque transmission, trajectory control, and human-robot collaboration, can to a large extent be overcome. Thereby, achieving dependable usage of such robots marks a key step towards utilizing their unique high safety features for physical human-robot collaboration in industrial applications. In order to get an insight into the safety capabilities as a collaborative robot, a risk analysis according to the new ISO/TS 15066 has been performed comparing the class of tendon driven ultra lightweight series elastic robots with downscaled and stiff robots. Further, the influence of high compliance during contact situations has been demonstrated during a collision test. In lightweight and downscaled robotic structures, the suitable joint position sensor size is limited, resulting in a relatively coarse discrete position signal. Furthermore, joint elasticities facilitate fast and oscillating motions containing a broad bandwidth of frequencies and velocities. In order to effectively damp the controlled motion and also for model-based computations, an accurate velocity estimation is essential. Kalman filter approaches already showed accurate performance in position signal based velocity estimation, but within a small bandwidth that correlates with the filter's measurement variance parameter. In order to create an estimation approach more suitable for the considered concept of tendon driven series elastic actuation, the velocity and frequency dependent optimal measurement variance parameter has been analyzed. The observed parameter characteristics serve as the basis for the proposed new measurement variance update rule. The resulting novel adaptive Kalman filter approach adapts better to the investigated application than the compared state-of-the-art approach. It produces a smoother and more accurate velocity estimation, as will be demonstrated in virtual and real robot experiments. In biologically inspired mechanical structures, tendons are used to transmit forces along a kinematic chain, which must satisfy high robustness requirements by simultaneously allowing small pulley radii. Cables or belts are often not applicable due to size and force requirements in contrast to thin synthetic fiber ropes. Besides the breaking force, only a little information about the ropes is provided by the manufacturer. Thus, further research is required to investigate whether a rope is suitable to be used as a reliable component of a drive train which mainly depends on its elongation behavior. In this regard, new systematic creep experiments regarding different materials, manufacturers, and diameters, as well as bending experiments are presented in this work. The findings obtained for the rope characteristics support the material selection decisions during the system design process and give insights into the long-term behavior. In order to monitor the long-term behavior, an observer approach is presented which does not need joint torque measurements but nevertheless enables elongation detection in tendon driven kinematic chains even subject to model inaccuracies. This has been demonstrated in simulated and real robot experiments. In an industrial application, a robotic system is expected to perform a motion without alterations over time accurately. The inherent uncertainties of the elastic tendon driven actuation can strongly influence the control performance and, thus, the reliability of the robot motion. State-of-the-art control approaches for joint elastic robot arms rely on accurate joint torque measurements and accurate models of drive train dynamics. Available torque sensors with a suitable torque range are heavy and large and do not fit into the lightweight actuation. Thus, so far only erroneous model-based joint torque estimation and consequently inaccurate load dependent friction models can be used, for which these control approaches perform unsatisfactorily. With the aim to reduce the model dependency, the performance of a friction observer under joint torque estimation errors is investigated in this thesis. This leads to the final proposed controller design that is able to compensate for both, drive train and load dynamics model inaccuracies. In particular, it integrates a friction observer into a state space approach without the need for explicit joint torque measurements eliminating the steady-state control errors and realizes contact handling. The resulting control performance has been evaluated in simulated and real robot experiments demonstrating an accurate and damped motion performance. An essential part of human-robot collaboration consists of physical interaction for joint task solving. To perceive physical interactions, the robotic arm must be able to estimate external forces. This can be achieved by disturbance observers, which only use internal data and are therefore suitable for the regarded system. In this thesis, the accuracy limitations are investigated and discussed theoretically and via simulated robot experiments to reveal the capabilities of tendon driven joint elastic robots regarding external torque estimations. Furthermore, the user should be able to teach the needed robot motion for the executed process without high effort. For this purpose, a teach-in controller that enables hand guided motions even under drive train model errors is presented and evaluated in realistic simulation. Besides the physical interaction, successful collaboration requires clear communication between the human and robot. To investigate the influence of robot state visualizations on the user's situational awareness and robot’s intention estimation in collaborative task solving, a preliminary user study is conducted. Further, metrics for wear estimation are introduced and discussed to support the human during the assessment of the robot's integrity and serve to increase trust in collaboration. |
||||
Alternatives oder übersetztes Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-76375 | ||||
Sachgruppe der Dewey Dezimalklassifikatin (DDC): | 000 Allgemeines, Informatik, Informationswissenschaft > 004 Informatik 600 Technik, Medizin, angewandte Wissenschaften > 620 Ingenieurwissenschaften und Maschinenbau |
||||
Fachbereich(e)/-gebiet(e): | 20 Fachbereich Informatik 20 Fachbereich Informatik > Simulation, Systemoptimierung und Robotik |
||||
Hinterlegungsdatum: | 12 Aug 2018 19:55 | ||||
Letzte Änderung: | 06 Dez 2018 10:11 | ||||
PPN: | |||||
Referenten: | von Stryk, Prof. Dr. Oskar ; Wollherr, PD Dr. Dirk | ||||
Datum der mündlichen Prüfung / Verteidigung / mdl. Prüfung: | 16 August 2017 | ||||
Export: | |||||
Suche nach Titel in: | TUfind oder in Google |
Frage zum Eintrag |
Optionen (nur für Redakteure)
Redaktionelle Details anzeigen |