IROS2019国际学术会议论文集2600

Novel lockable and stackable compliant actuation unit for modular +SPEA actuators Glenn Mathijssen1, Rapha el Furn emont1, Elias Saerens1, Manolo Garabini2, Manuel G. Catalano3 Dirk Lefeber1, Antonio Bicchi2,3and Bram Vanderborght1 AbstractOn compliant robotic systems, modularity is re- cently adopted for the ease of up- and downscaling and the possibility to downgrade the costs, by moving towards the combination of standard units instead of custom designs. However, modularity on the actuator level itself lacks more thorough evaluation. We have developed a novel lockable and stackable compliant actuation unit which can be used to form modular series-parallel elastic actuators (+SPEA). This paper describes the modular +SPEA layer architecture and discusses its two-way overrunner and rubber springs in detail, while providing experimental validation on each component as well. First experiments show the layer can deliver up to 20Nm. Finally, we present how a manipulator can be equipped with the modular +SPEA layers. I. MODULARITY AT THE ACTUATOR LEVEL Modularity in robotics is an increasing popular topic. The main motivations are robustness and increased resilience for system failure, ease of up- and downscaling, and the possibility of downgrading costs by moving away from custom designs towards a combination of standard units. On the robotics system level, multiple examples exist both in research and commercial fi elds, where modularity is key to the functionality of the system. Drone swarms and robot swarms, for instance 1 2, are a popular research and commercial topic nowadays. Modularity in social robotics has been introduced and described by 3 as well in order to create various platforms with a limited number of modular units. Compliant actuators are introduced by 4 with series elastic actuators (SEA) and later variable impedance actuators (VIA) as summarized in the review 5. They offered intrinsic safety 6 and robustness while offering the potential to increase the energy effi ciency such as done by 7. The series-parallel elastic actuators schematics, both the iSPEA 8 and +SPEA 9, offer the possibility to further increase the energy effi ciency by variably recruiting and locking parallel springs by one or more motors in parallel. This work was supported by ERC grant 337596 SPEAR. Elias Saerens is funded by PhD Fellowship of the Research Foundation - Flanders (FWO). 1Mathijssen, Furnemont,Saerens,LefeberandVanderborght arewithDepartmentofMechanicalEngineering,VrijeUni- versiteitBrussel,1050Elsene,BelgiumandFlandersMake glenn.mathijssenvub.ac.be 2Garabini and Bicchi are with Centro di ricerca E. Piaggio e dipartimento di ingegneria dellInformazione, University of Pisa, 56122, Pisa, Italy manolo.garabini 3Catalano and Bicchi are with Istituto Italiano di Tecnologia, Genova GE 16163, Italymanuel.catalanoiit.it The fi rst and second authors contributed equally to this work and should each be considered fi rst author. Digital Object Identifi er (DOI): see top of this page. (a) One +SPEA modular layer, combined to four +SPEA modular layers combined. (b) The stacked 4 +SPEA modular layers indicated in the use case of a manipulator arm. Fig. 1.The standardized +SPEA actuation layers can be combined to form an actuator. Visualization of a manipulator arm equipped with modular +SPEA layers. Here the layers are mounted on a base frame. Each joint is equipped with a different number of layers, from wrist over elbow to shoulder this is respectively twice 2, 3 and 4. The robot links in this visualization are 1.3m long. ModularcompliantactuatorssuchastheVSA-cube actuators from University of Pisa 10, offer the possibility to modularly build a robotic system. Each cube represents an actuator which can drive a joint. By linking several joints together, a robotic leg or arm can be produced. By means of dedicated electronics in each module, the modules can be daisy chained and easily linked together. Discrete muscle-like actuators, such as described in 11 12 13, consist of modular linear actuation units which can be combined in series and parallel to create an actuator. In sectionII, we discuss modularity on the actuator level, i.e. we aim to devise modular units which can be stacked to form an actuator. By combining units in series and parallel, the characteristics and performance of the actuator will change. Most often, robotic actuators are custom designs or are chosen from a line of modules. The novelty in this work is to design and explore stackable modular actuation layers which can then be combined together for a +SPEA to drive a single output axis. Much like the human body in which several muscles in parallel can drive a single joint. In this IEEE Robotics and Automation Letters (RAL) paper presented at the 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) Macau, China, November 4-8, 2019 Copyright 2019 IEEE Fig. 2.One modular +SPEA layer (actuator size 422mm by 149mm). work, we study the architecture of the +SPEA modular layers which we have designed, with specifi c focus on the two-way overrunning device used to lock recruited rubber springs. As such, the fi rst stackable and passively lockable +SPEA layer is presented in this work. In sectionIII we present the fi rst experiments, in sectionIV we present and describe a manipulator driven by +SPEA modular layers, and fi nally in sectionV we discuss our conclusion and future work. II. MODULAR +SPEA LAYER ARCHITECTURE A visual representation is shown in Fig.1. One modular +SPEA layer is a series elastic actuator SEA with a non backdrivable mechanism (i.e. a single input single output system (SISO). By stacking multiple layers in parallel as shown in Fig.1a, a +SPEA actuator is formed (i.e. a multiple input single output system (MISO). The design requirements are in the modular +SPEA layer design are fi vefold: 1) Comply with the +SPEA paradigm and thus install a non-backdrivable mechanism. More specifi cally, we aim to have a passive bi-directional highly effi cient component. 2) Allow a way to stack layers in parallel. 3) Minimize one dimension of the layer. 4) Strive for a lightweight and energy effi cient system. 5) Minimizing the need for complex custom designs. In the following subsection II-A we will discuss the modular +SPEA outline, after which we detail the non- backdrivable mechanism, the rubber springs and dyneema cables in sections II-B to II-D. A. Drivetrain and layer layout The modular +SPEA layer is shown in Fig.2. The actuator is composed of the elements listed here underneath. Since design simplicity was one of the clear goals, we have opted to minimize the number of custom parts and maximize the number of standard and semi-standard off-the-shelf parts. The mechanism consists of only 12 different custom parts. An overview of the actuator lay-out can be found in Fig.2, the main parts are listed here underneath: An electric motor equipped with an encoder. We have opted for the Maxon EX motor - 4 pole 22 (ref. 311536). The gearbox is Maxon Gear GP22 HP (ref. 370784) and encoder Maxon HEDL (ref. 110512). The two-way overrunner from the Bosch article number 3603JB3000, Pos. 45. The parts are then mounted in the custom designed parts. In line with the two-way overrunner a ball-screw of Misumi is installed (Misumi BSSC0802-207-SGC5), as such the two-way overrunner is positioned between two gear trains which limits the infl uence of the backlash. The bearing of the ball-screw is supported on the outside and fi xed on the inside. The ball-screw nut is connected to a traveler which is supported by two brass bushings (SPB-101410) with rubber oil impregnated patches, which slide along 2 support bars. The traveler is connected to the springs. Either directly to the spring connectors, by means of a hardened steel pin, or by means of the Dyneema cords which are in turn connected with a hardened steel pin. Finally, the Dyneema cables are connected to the output drum by mounting their end-loop around a bolt. The output drum is connected to the output shaft, which can in turn be connected to the output drum of the next layer. The size and weight of one layer is: The length is 422mm, which is close to the length of one link of the human arm, or the length of the link of a robotic co-worker. The length can be increased, to for example benefi t an increased output range. The height is 149mm. This can be chosen at will and upon design iterations it is estimated this can be reduced to 100mm. The depth is 40mm. This is minimized to stack different layers. Via design iterations, it is estimated this can be reduced to 20mm, which is the diameter of the motor. The weight of a single layer is currently 2.6kg. Via design iterations and optimization, it is estimated this can be reduced to 1.6kg. Finally the (theoretical) performance actuator characteris- tics of one layer are: Output torque: 30Nm Output range: 200o Torque bandwidth: 2Hz (based on actuator maximum speed and when considering a full cycle of max. counter-clockwise to max. counter-clockwise torque, and reverse, during a blocked output experiment). Output stiffness: 17.1 Nm rad (when considering a lin- earized spring stiffness). B. Non-Backdrivable Mechanism: Two-way Overrunning clutch The+SPEAparadigmischaracterizedbyanon- backdrivable element in each layer in order to keep the stored energy in the elastic element. The active magnetic (a) Forces acting on the wedgedroller(yellow) while blocking an output torque o. Adopted from 16. (b) Inner view of the two-way overrun- ning clutch. Two-way overrunning clutch test bench with a DC motor as input (the hammer driving the brass rollers is colored red) and a load drum as output. Fig. 3.Two-way overrunning clutch tests. The clutch was designed and produced in-house and 19mm in diameter. brakes used in 9 are off-the-shelf and thus easy to implement. On the other hand, however, they need to be controlled actively and can only be activated once the motor driving its layer came to a standstill since they cannot dissipate energy. Through the methodology and diagram in the overview paper 14 on locking mechanisms in robotics, a suitable locking mechanism is found: the two-way overrunning clutch as depicted in Fig.3. Overrunning clutches have an inner and outer raceway similar to bearings, with cylinders or balls (rollers, in yellow) between the two raceways and a wedge (in green) on one side. The two-way version, as described in this section, is altered to allow the input to drive the load counter-clockwise and clockwise by adding an input hammer (in red). As such, this clutch: Is passive: i.e. no control signal is needed to activate or deactivate. Is bi-directional. Has minimal losses from input to output, while it blocks once the output drives the input. Is small, lightweight, and can be found as a commercial component as for improved robustness and durability. A disadvantage of an overrunning clutch is the backlash which is between 10oand 50o, based on the experiments performed. As shown by 15, this can be minimized to 6o through design optimization. Moreover, its design requires high tolerance production techniques (up to 10m). 1) Design rules two-way overrunner: The design of two- way overrunning mechanisms and their design rules are not generalized yet. The basics can be found in 15 17. A brief summary here underneath. As shown in Fig.3a, it is assumed that a torque ois applied to the output of the two-way overrunner. The output torque is blocked by n rollers, which means the mechanism consists of 2n rollers in total to be able to block output torque in both directions. The output ring is assumed to be circular while the wedge (in green) is assumed to be straight. Fig. 4.Bi-directional experiments on the in-house designed and produced two-way overrunner with 3.66mm brass rollers, resulting in a locking torque of up to 3.5Nm. Different colors indicate different trials and 100% is reached once the rollers slip. Line r is perpendicular to the line connecting the centre of rotation (denoted with a cross) and the contact point of the ring and the roller. The torque oresults in the forces Fo and Frwhich can respectively be radially and tangentially decomposed in Fr,r, Fr,t, and Fo,rand Fo,t. The friction forces between the rollers and the output, and the rollers and the ring, are crucial to keep the rollers in the wedge. Via the locking condition a maximum for theta can be found, since the arctan() cannot be smaller than the angle 2. Otherwise, the tangential friction force cannot counter the output torque any more. Theoretically, ocan take any value, and can thus be infi nitely big. However, due to plastic deformations and thus a varying friction coeffi cient, the o is limited. 2) Experiments two-way overrunner: The experiments with the in-house designed and produced set-up in Fig.3 have shown a locking torque of up to 1Nm with roller diameter 3.56mm, and 3Nm with roller diameter of 3.66mm. The latter are shown in Fig.4. The tests were carried out by manually turning a bi-directional in-line torque sensor which transmitted the generated manual torque to the output of the two-way overrunning mechanism. The experiments were ended once the rollers started slipping. We did experience an increased resistance, and thus a decreased drive effi ciency, when turning the input with the 3.66mm brass rollers com- pared to the 3.56mm rollers. The tests taught us that tolerances are crucial in the mech- anism. Moreover, besides the and relation there are no clear design rules for each component. Therefore, the modular +SPEA actuator described in the following sec- tions was constructed with overrunning mechanisms found in commercial high-end drills (e.g. Bosch, Makita). We have opted for the two-way overrunner in Bosch article number 3603JB3000, Pos. 45. The backlash in this two-way overrunner is rather high, up to 40o. C. Rubber springs Compared to standard metal helical springs, rubbers have an excellent energy to weight ratio and strain rate. The disadvantage of rubbers is that their stress-strain behavior is non-linear and dependent on the history of the strain energy. As described by 18, rubbers are challenging materials for design engineers due to the following three properties: Cyclic property changes. Large deformation response. Non-linearity of the stress-strain behavior. We have opted for rubbers from Sandow Technic, which are often used as bungee cord. The main challenge is the interconnection between the rubbers and the system. We have opted to work with hydraulic hoses LH 3/8 U8000-06 which are then crimped onto a custom made connector. As shown in Fig.6, the inner part of the rubber strands are made out of bundled thin rubber held together by a woven fabric. In order to mount the perfectly fi tting hoses, the rubber is pressed onto a clamp which pushes the inner strands after inversion of the woven fabric. By cutting the thin rubbers and releasing the clamp, the hose can be mounted. A hydraulic press then fi xes the hoses on the rubbers. Whether deployed in constant or cyclic deformations, the stress-strain properties of rubber will vary. We performed fi rst basic test-bench experiments (on an Instron 5900R) as shown in Fig.7. In Fig.7a the difference between the very fi rst extension cycle and the last cycle is clearly shown. The majority of the cycles however shows a similar hysteresis pattern. The 18mm rubbers tested in Fig.7c can be used between 20% and 80% extension. Before 20% the behavior is strongly non-linear, past 80% the behavior is non-linear as well and moreover the fabric starts to stretch which highly increases the force-strain slope (as shown in Fig7a as well). The 18mm rubber tests were performed at 0.05Hz. The 18mm rubbers were tested till failure, which was at approx. 2500N. The springs failed because the hose and rubber disconnected on one side. D. Dyneema In order to provide a strong though lightweight connec- tion to the spring connectors, a cord solution was chosen. Different types of cords exist which could be useful, though the rope knotting usually is a challenge. As explained by 19, Dyneema cables with Brumel-lock splicing are an optimal solution to provide cords with two looped ends, as shown in Fig.2, without reducing the strength of the cord. Further information with regard to the Brumel-lock splicing is considered out of the scope of this paper, though can be found online fl uently. For the modular +SPEA, a 2mm Dyneema cord is used with a maximum force equal to 4060N (which is more than double the forces generated by the rubber springs). III. EXPERIMENTS The test-bench in Fig.5 used for the experiments consists of a Maxon load motor which can simulate any output scenario on the output shaft of the modular +SPEA layer under test. The in-line torque sensor (DRBK-200 200 N, ETH messtechnik) ensures we can measure the output torque while the load motor is equipped with an encoder as well to measure the output angle. The test bench is driven by a real-time data-acquisition (DAQ) system based on National Instruments input/output DAQ boards. PCI-6602 for the encoder readings and PCI-6229 for the analog input-outputs and digital outputs. The DAQ boards are installed on a PC (CoreTM2 CPU 6600 at 2.40 GHz, Intel) running Real- Time Windows Target and Simulink. The Maxon motor is driven by a commercial ESCON drive using reference inputs generated by the DAQ system. The full system was powered by regulated industrial power supplies (CP-E 24.0 V / 20 A, CP-E 48.0 V / 10 A, ABB). The aim of the experiments is to show the actuator performs as expected, and to determine the actuator performance in general. A. Non back-drivability characterization In order to measure and quantify the functionality of the two-way overrunning mechanism, we perform an identical experiment twice: once with the rollers