- Email: [email protected]
Time Domain Control for Passive Variable Motion and Force Scaling in Delayed Teleoperation
15th IFAC Workshop on Time Delay Systems 15th IFAC Workshop on Time Delay Systems Sinaia, Romania, September 2019 15th IFAC Workshop on Time Time9-11, Delay Systems 15th IFAC Workshop on Delay Systems Available online at www.sciencedirect.com Sinaia, Romania, September 9-11, 2019 15th IFAC Workshop on Time9-11, Delay Systems Sinaia, Romania, September 2019 Sinaia, Romania, September 9-11, 2019 Sinaia, Romania, September 9-11, 2019
ScienceDirect
IFAC PapersOnLine 52-18 (2019) 31–36
Time Domain Control for Passive Variable Time Time Domain Domain Control Control for for Passive Passive Variable Variable Motion and Force Scaling in Delayed Time Domain Control for Passive Variable Motion and Force Scaling in Delayed Motion and Force Scaling in Delayed Motion andTeleoperation Force Scaling in Delayed Teleoperation Teleoperation Teleoperation M. Panzirsch ∗∗ R. Balachandran ∗∗
M. Panzirsch ∗∗ R. Balachandran ∗∗ M. M. Panzirsch Panzirsch ∗ R. R. Balachandran Balachandran ∗ M. Panzirsch R. BalachandranAerospace Robotics and Mechatronics, Robotics and Mechatronics, German German Aerospace Center Center Robotics and and Mechatronics, German Aerospace Center Center Wessling, Germany (e-mail: [email protected]). Robotics Mechatronics, German Aerospace Wessling, Germany (e-mail: [email protected]). Institute of Wessling, Robotics and Mechatronics, Aerospace Center (DLR), 82234 Wessling, Germany (e-mail: German [email protected]). (DLR), 82234 Germany (e-mail: [email protected]). (DLR), 82234 Wessling, Germany (e-mail: [email protected]). Abstract: Abstract: Scaling Scaling of of motion motion and and forces forces has has always always been been of of high high relevance relevance in in teleoperation teleoperation setups setups Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Teleoperation setups are often affected by a delay in the communication channel. Most state since it allows the adaptation of workspaces of master and slave devices or to increase precision. Teleoperation setups are oftenof affected by aofdelay in and the slave communication channel. Most state since itart allows the adaptation workspaces master devices orbased to increase precision. Teleoperation setups are often oftenthat affected by aa stability delay in despite the communication communication channel. Most state of the control approaches guarantee delay are on the passivity Teleoperation setups are affected by delay in the channel. Most state of the art control approaches that guarantee stability despite delay are based on the passivity Teleoperation setups are often affected by a delay in the communication channel. Most state of the the art artwhich control approaches that guarantee guarantee stability despite delay arepaper basedproposes on the the passivity passivity criterion is highly restrictive to standard scaling methods. This different of control approaches that stability despite delay are based on criterion which is highly restrictive to standard scalingdespite methods. This paper proposes different of the art control approaches that guarantee stability delay are based on the passivity criterion which is highly restrictive to standard scaling methods. This paper proposes different time domain control concepts that the force the flow criterion which is highly restrictive to standard scalingor methods. This based paper on proposes different time domain control concepts that regulate regulate the motion motion force scaling scaling based the energy energy flow criterion which is highly restrictive toapproach standard scalingor methods. This motion paper on proposes different time domain control concepts that regulate the or force scaling based on the flow in delayed teleoperation systems. focuses on setups with down-scaling and time domain control concepts thatThe regulate the motion motion or force scaling based on the energy energy flow in delayed teleoperation systems. The approach focuses on setups with motion down-scaling and time domain control concepts that regulate the motion or force scaling based on the energy flow in applicable delayed teleoperation teleoperation systems. The impedance approach focuses focuses onThe setups with control motion is down-scaling anda is to variable motion and scaling. scaling integrated in in delayed systems. The approach on setups with motion down-scaling and is applicable to variable motion and impedance scaling. The scaling control is integrated in in delayed teleoperation systems. The approach focuses onThe setups with motion down-scaling andaaa is applicable applicable to time variable motion and impedance scaling. The scaling control is integrated in in state of the art delay control concept and its performance is analyzed in experiments. is to variable motion and impedance scaling. scaling control is integrated state of the art time delay control concept and its performance is analyzed in experiments. is applicable to time variable motion impedance The scaling controlin integrated in a state of delay control concept and performance is state of the the art art time delay controland concept and its itsscaling. performance is analyzed analyzed inisexperiments. experiments. Copyright 2019. Thedelay Authors. Published by Elsevier All rights reserved. state of the© art time control concept and itsLtd. performance is analyzed in experiments. Keywords: Keywords: Scaling, Scaling, Time Time Delay, Delay, Teleoperation, Teleoperation, Time Time Domain Domain Control Control Keywords: Keywords: Scaling, Scaling, Time Time Delay, Delay, Teleoperation, Teleoperation, Time Time Domain Domain Control Control Keywords: Scaling, Time Delay, Teleoperation, Time Domain Control 1. INTRODUCTION A large set of 1. INTRODUCTION A large set of control control concepts concepts for for delayed delayed teleoperation teleoperation 1. A based large set set of control concepts for delayed teleoperation is on an energy criterion. Colgate (1991) and 1. INTRODUCTION INTRODUCTION A large of control concepts for delayed teleoperation is based on an energy criterion. Colgate (1991) and Itoh Itoh 1. INTRODUCTION A large set control concepts for delayed teleoperation is based based on of anchose energy criterion. based Colgate (1991) and Itoh Itoh et al. (2000) a passivity design considering is on an energy criterion. Colgate (1991) and et al. (2000) chose a passivity based design considering is based on an energy criterion. Colgate (1991) and Itoh et al. (2000) chose aaInpassivity based design considering aa scattering matrix. the system of Itoh et al. (2000), et al. (2000) chose passivity based design considering Teleoperation is a mature technology that has a large scattering matrix. Inpassivity the system of design Itoh et considering al. (2000), al. (2000) chose a based Teleoperation is a mature technology that has a large et a scattering matrix. In the system of Itoh et al. (2000), the scaling design was specific in that it was separated a scattering matrix.was In specific the system of Itoh et al. (2000), Teleoperation is technology has aa large variety of application application fields ranging ranging fromthat nuclear research scaling design in that it was separated Teleoperation is aa mature mature technology that has research large the matrix.was In specific the system of Itoh etmethod al. (2000), variety of fields from nuclear thescattering scaling design was specific in that that it was was separated Teleoperation is medical a mature technology that has a large athe from the telemanipulator. The wave variables was scaling design in it separated variety of application fields ranging from nuclear research to space and the sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was variety of application fields ranging from nuclear research the scaling design was specific in that it was separated to space and the medical sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was variety of application fields ranging from nuclear research applied by Boukhnifer and Ferreira (2006). Secchi et al. from the telemanipulator. The wave variables method was to space and the medical sector. Some prefer an up-scaling of the master motions for a large slave robot workspace, applied by Boukhnifer and Ferreira (2006). Secchi et al. to space and the medical sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was of the master motions forsector. a large slave robot workspace, applied by Boukhnifer and Ferreira (2006). Secchi et al. to space and the medical Some prefer an up-scaling (2005) designed a scaled teleoperation setup with the portapplied by Boukhnifer and Ferreira (2006). Secchi et al. of the master motions for a large slave robot workspace, others as surgical applications require microor nano(2005) designed a scaledand teleoperation setup with the et portof the master motions for a largerequire slave robot workspace, applied by Boukhnifer Ferreira (2006). Secchi al. others as surgical applications microor nano(2005) designed a scaled teleoperation setup with the portof the master motions for a large slave robot workspace, Hamiltonian system representation. Often, it is assumed (2005) designed a scaled teleoperationOften, setup with the portothers as surgical applications require microor nanomanipulation capabilities in the extreme case (Onal and Hamiltonian system representation. it is assumed others as surgical applications require microor nano(2005) designed a scaled teleoperation setup with the portmanipulation capabilities in the extreme case (Onal and Hamiltonian systemare representation. Often, it is is assumed assumed others as surgical applications or downnanothat environments passive et Secchi system representation. it manipulation capabilities the extreme case (Onal and Sitti (2009)) which can be bein achieved by micromotion that environments are passive (Itoh (Itoh Often, et al. al. (2000), (2000), Secchi manipulation capabilities in achieved the require extreme (Onal and Hamiltonian Hamiltonian system representation. Often, it is assumed Sitti (2009)) which can by aacase motion downthat environments are passive (Itoh et al. (2000), Secchi manipulation capabilities in the extreme case (Onal and et al. (2005), Onal and Sitti (2009)) which presents a clear that environments are passive (Itoh et al. (2000), Secchi Sitti (2009)) which can be achieved by a motion downscaling. Apart from motion scaling also force scaling can be et al. (2005), Onal and Sitti (2009)) which presents a clear Sitti (2009)) which can be achieved by a motion downthat environments are Sitti passive (Itoh et al.For (2000), Secchi scaling. Apartwhich from motion scaling alsoby force scaling can be et al. (2005), Onal and (2009)) which presents a clear Sitti (2009)) can be achieved a motion downlimitation for the teleoperation scenario. example, a et al. (2005), Onal and Sitti (2009)) which presents a clear scaling. Apart from motion scaling also force scaling can be helpful, for example, in teleoperation training or for task limitation forOnal the teleoperation scenario. For example, a scaling. Apart from motion scaling alsotraining force scaling can be et al. (2005), and Sitti (2009)) which presents a clear helpful, for example, in teleoperation or for task limitation for the teleoperation scenario. For example, a scaling. Apart from motion scaling also force scaling can be beating heart in a medical scenario or a human interacting limitation for in the teleoperation scenario. For example, a helpful, for example, in teleoperation training or for task allocation in multilateral setups (Panzirsch et al. (2018)). beating heart a medical scenario or a human interacting helpful, for example, in teleoperation training or for task limitation for the teleoperation scenario. For example, a allocation in multilateral setups (Panzirsch et al. (2018)). beating heart in a medical scenario or a human interacting helpful, for example, in teleoperation training or for task with a robot in an ambient assisted living scenario presents beating heart in a medical scenario or a human interacting allocation in multilateral setups (Panzirsch et al. (2018)). Depending on the combination of motion and force scaling with a robot in an ambient assisted living scenario presents allocation in multilateral setups (Panzirsch et al. (2018)). beating heart in a medical scenario or a human interacting Depending on the combination of motion and force scaling with a robot in an ambient assisted living scenario presents allocation inon setupsof (Panzirsch etforce al. (2018)). an active environment. Also, several control approaches a robot in an ambient assisted living scenario presents Depending the combination motion scaling factors, so-called power scaling, impedance scaling or pure with an active environment. Also, several control approaches Depending onmultilateral thepower combination ofimpedance motion and and force scaling with a robot in an ambient assisted living scenario presents factors, so-called scaling, scaling or pure an active environment. Also, several control approaches Depending on the combination of motion and force scaling require power scaling settings (Boukhnifer and Ferreira an active environment. Also, several control approaches factors, so-called power scaling, impedance scaling or pure motion or force scaling can be designed. The works of require power scaling settings (Boukhnifer and Ferreira factors, so-called or pure active environment. Also, several control approaches motion or force power scalingscaling, can beimpedance designed. scaling The works of an require power scaling settings (Boukhnifer and Ferreira factors, so-called power scaling, impedance scaling or pure (2006)) since it is an intrinsically passive functionality. But require power scaling settings (Boukhnifer and Ferreira motion force scaling can works Colgate (1991), Vander Poorten etdesigned. al. (2006) (2006)The and Goldfarb Goldfarb since it scaling is an intrinsically passive functionality. But motion or or forceVander scalingPoorten can be beet designed. The works of of (2006)) require power settings (Boukhnifer and Ferreira Colgate (1991), al. and (2006)) since it is an intrinsically passive functionality. But motion or force scaling can be designed. The works of the power scaling values differ extremely from impedance (2006)) since it is an intrinsically passive functionality. But Colgate (1991), Vander Poorten et al. (2006) and Goldfarb (1998) discuss different force scaling designs according to the power scaling values differ extremely from impedance Colgate (1991), Vander Poorten et al. (2006) and Goldfarb (2006)) since itwhich is an intrinsically passive functionality. But (1998) discuss different force scaling designs according to the power scaling values differ extremely from impedance Colgate (1991), Vander Poorten et al. (2006) and Goldfarb scaling values are often preferable. the power scaling values differ extremely from impedance (1998) discuss different force scaling designs according to the properties of the the environmental environmental impedance that can can values which are often preferable. (1998) discuss different force scaling designs according to scaling the power scaling values differ extremely from impedance the properties of impedance that scaling values which are often preferable. (1998) discuss different force scaling designs according to values which are often preferable. the properties of environmental that be primarily inertial, or exhibit exhibitimpedance viscous damping. damping. theprimarily propertiesinertial, of the the elastic environmental impedance that can can scaling Here, focus on passivity criterion values often preferable. be or viscous Here, we we focuswhich on the theare passivity criterion since since most most state state the propertiesinertial, of the elastic environmental impedance that can scaling be primarily elastic or exhibit viscous damping. be primarily inertial, elastic or exhibit viscous damping. Here, we focus on the passivity criterion since most of the art approaches for time-delayed teleoperation as Here, we focus on the passivity criterion since most state state Scaling factors have been applied within different teleopof the art approaches for time-delayed teleoperation as be primarily inertial, elastic or exhibit viscous damping. Here, we focus on the passivity criterion since or most state Scaling factors have been applied within different teleop- the of the art approaches for time-delayed teleoperation as wave variables method (Niemeyer (1996)) the time of the art approaches for time-delayed teleoperation Scaling factors have been applied within different teleoperation control architectures and with a variety of stability the wave variables method (Niemeyer (1996)) or the time Scaling control factors architectures have been applied within different teleop- of the art approaches for time-delayed teleoperation as as eration and with a variety of stability the wave variables method (Niemeyer (1996)) or the time Scaling factors have been applied within different teleopdomain passivity approach (TDPA, Ryu et al. (2010), the wave variables method (Niemeyer (1996)) or the time eration architectures and variety of stability stability domain passivity approach (TDPA, Ryu et al. (2010), analyzes. In Speich Speich and Goldfarb Goldfarb (2002), the transparency transparency eration control control architectures and with with aa variety of the wave variables method (Niemeyer (1996)) or the time analyzes. In and (2002), the domain passivity passivity approach (TDPA, Ryu et al. al. (2010), (2010), eration control architectures and with a with variety of stability domain Panzirsch et are based this The approach et analyzes. In Speich and (2002), the transparency of scaled position-force architecture loop-shaping Panzirsch et al. al. (2019)) (2019)) are (TDPA, based on on Ryu this criterion. criterion. The analyzes. In position-force Speich and Goldfarb Goldfarb (2002), the transparency domain passivity approach (TDPA, Ryu et al.concepts (2010), of aa scaled architecture with loop-shaping Panzirsch et al. (2019)) are based on this criterion. The analyzes. In Speich and Goldfarb (2002), the transparency benefit is the high modularity of energy-based Panzirsch et al. (2019)) are based on this criterion. The of a scaled position-force architecture with loop-shaping compensators has been evaluated. Also, the H approach benefit is the high modularity of energy-based concepts of a scaled position-force architecture with loop-shaping ∞ Panzirsch et al. (2019)) are based on this criterion. The compensators has been evaluated. Also,with the Hloop-shaping ∞ approach benefit is the high modularity of energy-based concepts of a scaled position-force architecture which allows, for example, uncomplicated extensions of benefit is the high modularity of energy-based concepts compensators has been evaluated. Also, the H approach (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and ∞ which allows, for example, uncomplicated extensions of approach compensators has been evaluated. Also, the H ∞ benefit is the high modularity of energy-based concepts (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and which allows, for example, uncomplicated extensions of approach compensators has been evaluated. Also, the H bilateral to multilateral setups (Panzirsch et al. (2013)). which allows, for example, uncomplicated extensions of ∞ (Yan and Salcudean Salcudean (1996), Boukhnifer et al. al. (2004)) (2004)) and bilateral to multilateral setups (Panzirsch et al. (2013)). a sliding mode control (Khan et al. (2009)) have been (Yan and (1996), Boukhnifer et and which allows, for propose example, uncomplicated extensions of a sliding mode control (Khan et al. (2009)) have been bilateral to multilateral setups (Panzirsch et al. (2013)). (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and In this paper, we a time domain control concept bilateral to multilateral setups (Panzirsch et al. (2013)). a sliding mode control (Khan et al. (2009)) have been implemented for scaled teleoperation. Power scaling was In this paper, we proposesetups a time(Panzirsch domain control concept a sliding mode control (Khan et al. (2009)) have been bilateral to multilateral et al. (2013)). implemented for scaled teleoperation. Power scaling was In this paper, we propose a time domain control concept a slidingbymode (Khan et al. (2009)) have been for passive scaling that be teleoperation paper, we propose a time domain in control concept implemented for scaled teleoperation. Power was applied Ferreira Boukhnifer for this passive scaling that can can be applied applied in teleoperation implemented forcontrol scaled and teleoperation. Power scaling scaling was In paper, we propose a time domain control concept applied by Boukhnifer Boukhnifer and Ferreira (2006), (2006), Boukhnifer for passive scaling that can be in teleoperation implemented for scaled teleoperation. Power scaling was In scenarios with active environments. In contrast to former for this passive scaling that can be applied applied in teleoperation applied by Boukhnifer Boukhnifer and Ferreira (2006), Boukhnifer et al. (2004) and Jazayeri and Tavakoli (2013). Jazayeri scenarios with active environments. In contrast to former applied by and Ferreira (2006), Boukhnifer for passive scaling that can be applied in teleoperation et al. (2004) and Jazayeri and Tavakoli (2013). Jazayeri scenarios with active environments. In contrast to former applied by Boukhnifer and Ferreira (2006), Boukhnifer approaches, the concept guarantees passivity in setups scenarios with active environments. In contrast to former et al. al.Tavakoli (2004) (2013) and Jazayeri Jazayeri and Tavakolistability (2013).for Jazayeri and focused on absolute scaled approaches, the concept guarantees passivity in setups et (2004) and and Tavakoli (2013). Jazayeri with active environments. Inpassivity contrast to former and Tavakoli (2013) focused and on absolute stability for scaled scenarios approaches, the concept guarantees passivity in setups et al. (2004) and Jazayeri Tavakoli (2013). Jazayeri with pure motion, pure force and impedance scaling, and approaches, the concept guarantees in setups and Tavakoli (2013) focused on absolute stability for scaled sampled-data systems. Impedance scaling has been conwith pure motion, pure force and impedance scaling, and and Tavakoli (2013) focused on absolute stability for scaled approaches, the concept guarantees passivity in present setups sampled-data systems. Impedance scaling has been con- with pure motion, pure force and impedance scaling, and and Tavakoli (2013) focused on absolute stability for scaled allows for time-varying scaling designs. Also, we with pure motion, pure force and impedance scaling, and sampled-data systems. Impedance scaling has been considered by Onal and Sitti (2009). Vander Poorten et al. allowspure for motion, time-varying scaling designs. Also,scaling, we present sampled-data systems. Impedance scaling Poorten has beenetconwith pure force and impedance and sidered by Onal and Sitti (2009). Vander al. allows for time-varying scaling designs. Also, we present sampled-data systems. Impedance scaling has been conhow the conservatism of time domain passivity control forconservatism time-varying ofscaling designs. Also, we present sidered performed by Onal Onal and and Sitti (2009). (2009). Vander Poorten et on al. allows (2006) an absolute stability analysis based how the time domain passivity control sidered by Sitti Vander Poorten et al. allows for time-varying scaling designs. Also, we present (2006) performed an absolute stability analysis based on howdelayed the conservatism conservatism ofmotion time domain domain passivity control sidered by Onal and Sitti (2009). Vander Poorten et on al. how in setups scaling can the time control (2006) performed an absolute absolute stability analysis based on the scattering matrix of fixed-scale teleoperin delayed setups with withof scaling passivity can be be reduced. reduced. (2006) performed an stability analysis based how the conservatism ofmotion time domain passivity control the scattering matrix of a a delay-free, delay-free, fixed-scale teleoperin delayed setups with motion scaling can be reduced. (2006) performed an absolute stability analysis based on in delayed setups with motion scaling can be reduced. the scattering matrix of a delay-free, fixed-scale teleoperation setup. the of a delay-free, fixed-scale teleoper- in delayed setups with motion scaling can be reduced. ationscattering setup. matrix the scattering matrix of a delay-free, fixed-scale teleoperation ation setup. setup. ation setup. 2405-8963 Copyright © 2019. The Authors. Published by Elsevier Ltd. All rights reserved. ∗ ∗ Institute of Institute of ∗ ∗ Institute of (DLR), 82234 of ∗ Institute (DLR), 82234
Copyright © 2019 IFAC 55 Copyright © under 2019 IFAC 55 Control. Peer review responsibility of International Federation of Automatic Copyright © 55 Copyright © 2019 2019 IFAC IFAC 55 10.1016/j.ifacol.2019.12.202 Copyright © 2019 IFAC 55
2019 IFAC TDS 32 Sinaia, Romania, September 9-11, 2019
Scaling ν
vm
ν vm + -
Human M aster Fh
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
vs
Ctrl
+ Fcµ
Slave
Env.
+ -
µ
Fe
Fc
Fig. 1. Signal flow diagram of a 2-Channel architecture with scaling. Agent Λ1
Fig. 3. Experimental setup: two 1-DoF rotational devices.
Agent Λ2
v1
v1
v2
v3
v3
+
+
+
+
+
F2
Ctrl F3
+ -
Human Fh M aster F1 l P ort 1
Scaling
-
-
P ort 2
P ort 3
Slave
Fe
device (Zm = µFc vm ). With vm = νvs , impedance scaling requires µ = 1/ν which is highly contradictive to the passive power scaling. The experiment in Fig. 5 presents an impedance scaling teleoperation with ν = 0.5 and µ = 2. The energy plot clearly shows the energy generation during the operation (negative slope of E2port ) when the slave leads the motion at t > 3.5s.
Env.
-
The impedance scaling focuses on the haptic perception of impedance at the master device and neglects the visual feedback. Alternatively, it can be argued that the human operator relates the force displayed at the master device to the motion of the slave robot (instead of the master motion). Then, a force scaling of µ = 1 is reasonable. This may be especially relevant for master and slave systems with similar workspace where a pure motion scaling is applied to increase the precision of the operation. In teleoperation scenarios with free motion and soft walls, the pure motion scaling promises a good perception of the environmental interaction whereas during hard contacts (low slave motion) the impedance scaling might be more intuitive. The reader is referred to the work of Goldfarb (1998) for more details on the choice of force scaling factors depending on the environment contact. As visible from the experiment in Fig. 6 and the energy plot E2port , the pure motion scaling (µ = ν, µ = 1, ν = 0.5) is not passive.
Fig. 2. Network representation of a 2-Channel architecture with scaling.
The method promises high transparency especially during motion down-scaling. The paper is structured as follows: Section 2 analyzes the energetic behavior of motion and force scalings in control networks. A power-based and an energy-based time domain passivity control for variable motion and force scaling are introduced in Section 3. The experimental evaluation which includes time delay teleoperation setups is presented in Section 4. Finally, Section 5 concludes the work. 2. PROBLEM STATEMENT The signal flow diagram of a scaled teleoperation setup and the respective network representation are depicted in Fig. 1 and Fig. 2 respectively. A Human uses a M aster input device to control a slave robot in its environment. The two devices can be coupled with a PD controller (Ctrl). The variables ν and µ scale the motion (velocity) and force respectively (ν, µ ∈ [0, ∞[). The following relations ν between the signals holds: v1 = vm , v2 = vm , v3 = v s , µ F1 = Fc , F2 = F3 = Fc . The network representation is an electrical diagram that allows the power observation at the ports i with power-conjugated signals force Fi and velocity vi . The Scaling as well as the Ctrl are 2-port networks.
Depending on the applications, also a variable velocity scaling design can be reasonable. For example, the velocity scaling can be a function of the master velocity (ν = f (vm )) such that for fast master motion ν = 1 since it can be assumed that the human wants to move far. Whereas at lower master motion, the velocity scaling can be faded down to ν = 0.5 since the human wants to command a precise motion.
In the following, a passivity control for arbitrary and variable scaling values is designed that leads to a passive scaling network subsystem that can be applied to passivity-based control frameworks in a highly modular manner.
Concerning an analytical analysis of a scaling network in the frequency domain (e.g. Raisbeck’s criterion), the scaling subsystem is only passive if the motion and force scaling values are equal (power scaling, µ = ν). That means that, for the sake of passivity, in case of motion up-scaling, the force feedback has to be scaled up and in case of motion down-scaling, the feedback force has to be scaled down. The experiment in Fig. 4 shows that with equal µ and ν (µ = ν = 0.5), the energy over the scaling 2-port E2port is always zero which confirms its passivity. The experiments were performed with two 1-DoF rotational devices (SENSODRIVE, compare Fig. 3). The coupling software was executed on a QNX host machine at a sampling rate of 1kHz.
3. PROPOSED TIME DOMAIN CONTROL APPROACH As it can be analyzed from the energy plots in Fig. 4 to Fig. 6, the scaling dissipates or generates (positive and negative slope of E2port respectively) energy depending on the scaling values and on which device leads the motion. The power dissipation and generation can be calculated as follows (if ν, µ = 0): ν < P1 , if ν < µ (case 1) (1) P2 = P1 ≥ P1 , if ν ≥ µ (case 2). µ
With the classical impedance scaling, the slave side impedance (Zs = Fc vs ) can be displayed at the master 56
−2 1
1.5
2
2.5
3
3.5
4
4.5
−0.5 −1 −1.5 1
1.5
2
2.5
3
3.5
4
4.5
5
1
3
4
5
−1 −1.5 1
2
3
4
5
6
1.5
2
2.5
3
3.5
4
4.5
µForce νVelocity
1 0.5 0 0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 2
4
5
1.5
2
2.5
3
3.5
4
4.5
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0.5
1
1.5
2
2.5
3
3.5
4
4.5
−1
0
6
µForce νVelocity
1 0 2
3
4
Time [s]
Time [s]
Fig. 4. Power scaling.
3
2
1
5
1
−0.5 1
5
0.5
−1.5
Scaling [−]
1
−2
−0.5
E2port [J ]
0.2
Scaling [−]
0.5
Master Slave
0
0.5
0.4
0
2
0.5 0
0.6
0.1
33
6
−0.5
E2port [J ]
E2port [J ]
2
0.5 0
0.2
Scaling [−]
Master Slave
−2
5
0.5 0
0.5
0
P osition [rad]
Master Slave
2
Torque [Nm]
0 −1
P osition [rad]
1
0.5
Torque [Nm]
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Torque [Nm]
P osition [rad]
2019 IFAC TDS Sinaia, Romania, September 9-11, 2019
Fig. 5. Impedance scaling.
5
6
1
µForce νVelocity
0.5 0 0.5
1
1.5
2
2.5
3
3.5
4
4.5
Time [s]
Fig. 6. Pure motion scaling.
that this adaptation is not problematic: In this work, we focus on a down-scaling of the velocity such that the teleoperated slave robot moves slower than the master device. Thus, the controller has to adapt (increase) the velocity scaling if energy is generated by the scaling. As analyzed before, the scaling generates energy if energy is flowing from slave to master, that means if the master device moves out of a wall contact or if the slave leads the motion. If the scaling is not varied (time domain control inactive), the master moves much faster than the slave, which might be dangerous if the slave leads the motion. In contrast, if the velocity scaling is increased in such situations (by the passivity control action), the master device moves as slow and as far as the slave which is a much safer procedure.
In case 1, energy is dissipated in the left to right energy flow direction (L2R) and generated in the right to left (R2L) energy flow direction. In case 2, it behaves vice versa. The power flow direction can be distinguished by the sign of the power Pi . The power values PiL2R and PiR2L are positive by definition: Pi , if Pi > 0 L2R Pi = (2) 0 , if Pi ≤ 0, −Pi , if Pi < 0 (3) PiR2L = 0 , if Pi ≥ 0. In typical teleoperation setups, the master moves the slave such that in case 1 (for example at pure motion downscaling) the scaling is mostly overall passive. Still, classical frequency-based stability approaches do not consider this power flow direction dependency and therefore a passive power scaling (ν = µ) has to be chosen or active environments are not allowed. Applying a time domain energy observation and control, this conservatism can be heavily reduced. In the following, different time domain control approaches for case 1 in (1) will be discussed that consider different power and energy criteria. Furthermore, they differ by the adaptation of motion or force scaling. Finally, the integration of the control approach in a state of the art time delay control approach is presented.
Now, two approaches that consider a power-based and an energy-based criterion respectively are presented. Power-based passivity criterion (Approach 1): As explained before, the power flow direction can be analyzed and the focused motion down-scaling produces energy only in the R2L flow direction, which means when the slave leads the master. Therefore, a passivity controller can be designed that adapts the scaling for the sake of passivity to passive power scaling (µ = ν) if power flows in R2L direction: The velocity scaling des ν , if P1L2R > 0 ν(k) = (4) µ , if P1L2R = 0, or alternatively, the force scaling can be adapted, therefore: des , if P1L2R > 0 µ (5) µ(k) = ν , if P1L2R = 0. Note that force scaling may lead to disturbances in the teleoperation setup and therefore, velocity scaling should be favoured. Still, for the sake of completeness, the force scaling is also presented here. Experiments concerning impedance and pure motion scaling are presented in Section 4.
3.1 Time domain control for passive scaling In the following section, we concentrate on case 1 (ν < µ) settings. The proposed control method preserves passivity through the online adaptation of the velocity or force scaling as soon as passivity is violated. Therefore, considering motion down-scaling, the controller has to increase the velocity scaling or to decrease the force scaling in phases of undesired energy generation to achieve ν = µ (passive power scaling). In general, the force scaling should remain constant during teleoperation since the forces are directly perceived at the master device. In contrast, the velocity adaptation is not as obvious and disturbing to the operator since the operator focuses on the position command which is the integral of the adapted variable. On the first sight, the up-scaling of the velocity by the controller may appear as an unintuitive solution. But, a closer analysis reveals
The negative aspect of the power-based controller is that the scaling is not only varied when the slave leads the motion but also when the master leaves a wall contact (P1L2R = 0). Note that if the ν adaptation is chosen, 57
2019 IFAC TDS 34 Sinaia, Romania, September 9-11, 2019
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Agent Λ1
Agent Λ2
v1
v1
v2
v3
v4
v5
v6
v6
+
+
+
+
+
+
+
+
F3 P C2 F4 Scaling F5
Ctrl F6
CC
+ -
Human Fh M aster F1 P C1 F2 l P ort 1
P ort 2
-
-
P ort 3
P ort 4
P ort 5
Slave
-
Fe
Env.
-
P ort 6
Fig. 7. Network representation of the TDPA (Ryu et al. (2010)) for delayed teleoperation with scaling. the slave leaves the wall contact faster. As soon as the wall contact is over (P1L2R > 0), the standard scaling is reactivated. For example, in pick and place tasks, the operator wants to feel a wall contact, but also wants to leave it as fast as possible. Then, the power-based velocity adaptation has no negative effect. In soft environments, on the other hand, the slave leaves the wall contact faster and the perception of the wall contact is altered. Still, this may not be critical since the impedance of an object is mainly analyzed in the penetration direction.
time domain passivity control (TDPA, Ryu et al. (2010)). The TDPA introduces passivity observers that measure the energy that has been generated by the CC and passivity controllers (PC) that dissipate energy to ensure the passivity of the two-port between ports 1 and 4. This approach also considers two directions of energy flow and therefore, two PCs are implemented. An impedance-type P C1 varies the force which is sent to the master side with a variable damping α that depends on the energy that has to be dissipated and on the velocity v2 at the P C1 port. An PC admittance-type P C2 varies the velocity (v4 = v3 − vm ) which is sent to the slave side with a variable damping β that is calculated from the energy that has to be dissipated and the force F3 at the P C2 port. Since this variation can be interpreted as a down-scaling ν P C
Energy-based passivity criteria (Approach 2): This negative aspect of the power-based controller motivates for another solution that circumvents the scaling adaption during one wall contact. In contrast to the power-based design (Approach 1), the following concept considers the potential energy storage in the coupling controller that results from the coupling controller’s spring-like element.
PC del ν P C = min(vm /vm , 1),
(calculated from the delayed master reference velocity del PC vm = v3 and the velocity output vm of the P C2), it can be considered for the desired velocity scaling ν ∗ that PC acts on vm in case of velocity scaling adaptation (4) or (6): 1 , if ν P C < ν des ∗ ν = (9) des PC min(1, ν /ν ) , if ν P C ≥ ν des ,
This energy storage Est is built up, for example, during a wall contact: Est (k) = Est (k − 1) + (P2L2R (k) + P3R2L (k) − P3L2R (k) − P2R2L (k))Ts , with the sampling time Ts (compare Fig. 2).
During a wall contact, energy is charged into the coupling controller’s spring. The same amount of energy is released when the penetration into the wall is reduced. Therefore, a time domain controller for passive scaling that considers the energy ESt instead of power does not vary the scaling during the wall contact (in contrast to Approach 1). The velocity scaling
such that the effective overall scaling ν ∗ ν P C of the dedel layed master velocity vm is not higher than ν des (if the time domain controller for passive scaling is not active, otherwise ν = µ). The consideration of the P C2 scaling in the desired motion scaling ν leads to a less conservative passivity control since the P C2 dissipation leads to a position drift (p4 = p3 ) after integration of v4 . If the PC dissipation would not be considered in ν = ν ∗ ν P C , the velocity would be further reduced and position drift increased. The consideration of a force scaling adaptation (5) or (7) in P C1 is not presented here.
ν des , if Est (k) ≥ 0 (6) µ , if Est (k) < 0, or alternatively, the force scaling can be adapted therefore: µdes , if Est (k) ≥ 0 (7) µ(k) = ν , if Est (k) < 0. ν(k) =
(8)
4. EXPERIMENTAL EVALUATION
To assure that the scaling switches when the leading role switches from master to slave or vice versa, the reference ∗ has to be reset when xm = xs (with a energy storage ESt certain threshold). If Est (k) = 0, the old values of ν and µ have to be set. In contrast to the power-based adaption (Approach 1), force scaling may lead to lesser disturbances in Approach 2 since the scaling is not adapted during a wall contact.
The first experiment (see Fig. 8) presents the power scaling method (Approach 1) with pure motion scaling (ν des = 0.5, µdes = 1) and velocity scaling adaptation (4). The master moves the slave device in free motion and against a wall (t = [2s − 6s]). The velocity scaling is adapted when the master leaves the wall penetration, since power is flowing from slave to master (P1L2R = 0). Afterwards (t = [6s − 7.5s]), the slave moves the master and the velocity scaling is set to 1 (P1L2R = 0). With this control strategy, a passive scaling can be achieved, as it is confirmed by the purely positive 2-port energy E2port of the scaling subsystem.
3.2 Scaling in Time Delay Setups Figure 7 presents the network representation of a delayed teleoperation setup with communication channel CC and 58
5
4
6
5
7
6
7
3
4
5
6
7
8
0.2 0.1 0 2
3
4
5
7 Force Velocity
6
0.5 0 2
3
4
5
6
7
8
2
2.5
3
3.5
4
4.5
5
5.5
0 −1 1.5 1.5
2
2.5
3
3.5
4
4.5
5
1
0 1.5 1
2
2.5
3
3.5
4
4.5
5.5
5
5.5
0.5 0 1.5 2
2
1
2.5
3
3.5
4
4.5
5
2
2.5
3
Time [s]
3.5
4
4.5
5
Master Slave
0 1
5.5
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
3
4
5
6
0
L2R R2L
5
0 0 0.1
0.05 0 0
5.5
µForce νVelocity
0 1.5
2 1 0 −1
−1 0
L2R R2L
0.5
8
1
P osition [rad]
P osition [rad] 8
L2R R2L
0.5 0 2
8
35
Power [J /s] Torque [Nm]
3
4
Master Slave
E2port [J ]
2 1
3
3 2 1 0 −1 1.5 1
Scaling [−]
Scaling [−]
E2port [J ] Power [J /s] Torque [Nm]
2 0.2 0 −0.2 −0.4
Master Slave
E2port [J ] Power [J /s] Torque [Nm]
2 1 0 −1
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Scaling [−]
P osition [rad]
2019 IFAC TDS Sinaia, Romania, September 9-11, 2019
1
µForce νVelocity
0.5 0 0
1
2
Time [s]
Time [s]
Fig. 8. App1: Power Control with pure mo- Fig. 9. App1: Power control with impedance Fig. 10. App1: Power control with vari-
0 −0.5 1
1.5
2
2.5
3
2
L2R R2L
1 0 0.5
1
1.5
2
2.5
E2port ∗ ESt
0 0.5
1
1.5
2
2.5
0.5 0 0.5
µForce νVelocity 1
1.5
2
2.5
3
3
3.5
4
4.5
5
5.5
Position [rad] 6
0
−2 6 1
2
2.5
3
3.5
4
4.5
5
5.5
6
L2R R2L
2 1 0 2
2.5
3
3.5
4
4.5
5
5.5
6
0.2
E2port ∗ ESt
0.1 0 2
2.5
3
3.5
4
4.5
5 µForce 5.5
6
νVelocity
1 0.5
Master Slave 8
10
12
14
16
8
10
12
14
16
10
12
14
16
10
12
14
16
10
12
14
16
12
14
16
0.5 0
−0.5 6 10
−1
3
1
2.5
−0.5
3
0.2 0.1
2 0.5
Scaling [−]
Scaling [−]
Energy [J ]
Power [J /s]
−1 0.5
P osition [rad]
3
L2R Energies Torque [Nm] [J ]
2.5
Scaling [−] Scaling [−]
2
0 −1
5
E2L2R E3L2R E4L2R
0 6 1
8
0.5 0 6 1
ν PC ν∗ 8
0.5
0 6 0.05
νVelocity 8
4 −5
1.5
Master Slave
E2port [J ]
1
Torque [Nm]
0.5
3 2 1 0 −1
Power [J /s]
0.5
Master Slave
able motion scaling and velocity scaling adaptation.
scaling and force scaling adaptation. textfill next line
Energy [J ]
P osition [rad]
2 1 0 −1
Torque [Nm]
tion scaling and velocity scaling adaptation. textfill next line
0 2
2.5
3
3.5
4
4.5
5
5.5
6
0 6
8
10
Time [s]
Time [s]
Time [s]
Fig. 11. App2: Energy control with motion Fig. 12. App2: Energy control with force Fig. 13. TDPA with Approach 1 and scaling and velocity scaling adaptation. textfill next line
velocity scaling adaptation at 100ms roundtrip-delay
scaling and force scaling adaptation. textfill next line
In the second experiment (see Fig. 9), an impedance scaling (ν des = 0.5, µdes = 2) with force scaling adaptation (5) is presented. The master moves the slave t = [1.5s − 4.3s] and the slave overtakes the lead at t = [4.3s − 6s]. Comparing the power and scaling plots, it is obvious that the force scaling is adapted when energy flows from slave to master (R2L direction, P1L2R = 0). It can be seen from the energy plot that also the force adaptation leads to a passive scaling 2-port.
∗
ν =
ν des
ν
, if |vm | < vmin + , if vmin ≤ |vm | < 2 rad m |−vmin s (1 − ν des ) v|vmax −vmin rad 1.2 , if |vm | ≥ 2 s .
des
The variable scaling was designed such that at fast motions, the velocity scaling is ν = 1.2. If ν > 1 (t = [4.8s − 5.2s]), the time domain scaling controller has to set the force scaling to µ = ν independent of the power flow to maintain passivity. High master velocities appear at free motions and therefore, the increase of force scaling has no unintuitive effect. If the controller has to act, the velocity scaling is set to ν = µ (t = [5.5s − 6.8s]). Again,
The third experiment in Fig. 10 presents a variable motion scaling and power-based velocity scaling adaptation (4). The velocity scaling depends on the master velocity ν = f (vm ):
59
2019 IFAC TDS 36 Sinaia, Romania, September 9-11, 2019
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Proceedings. 1998 IEEE International Conference on Robotics and Automation, volume 2, 1609–1614. IEEE. Itoh, T., Kosuge, K., and Fukuda, T. (2000). Humanmachine cooperative telemanipulation with motion and force scaling using task-oriented virtual tool dynamics. IEEE Transactions on robotics and automation, 16(5), 505–516. Jazayeri, A. and Tavakoli, M. (2013). Absolute stability analysis of sampled-data scaled bilateral teleoperation systems. Control Engineering Practice, 21(8), 1053– 1064. Khan, S., Sabanovic, A., and Nergiz, A.O. (2009). Scaled bilateral teleoperation using discrete-time sliding-mode controller. IEEE Transactions on Industrial Electronics, 56(9), 3609–3618. Niemeyer, G.D. (1996). Using wave variables in time delayed force reflecting teleoperation. Ph.D. thesis, Massachusetts Institute of Technology. Onal, C.D. and Sitti, M. (2009). A scaled bilateral control system for experimental one-dimensional teleoperated nanomanipulation. The International Journal of Robotics Research, 28(4), 484–497. Panzirsch, M., Artigas, J., Ryu, J.H., and Ferre, M. (2013). Multilateral control for delayed teleoperation. In Proc. of 2013 International Conference on Advanced Robotics, 1–6. Panzirsch, M., Balachandran, R.R., Weber, B., Ferre, M., and Artigas, J. (2018). Haptic augmentation for teleoperation through virtual grasping points. IEEE Transactions on Haptics, 11(3), 400–416. Panzirsch, M., Ryu, J.H., and Ferre, M. (2019). Reducing the conservatism of the time domain passivity approach through consideration of energy reflection in delayed coupled network systems. Mechatronics, 58, 58–69. Ryu, J.H., Artigas, J., and Preusche, C. (2010). A passive bilateral control scheme for a teleoperator with timevarying communication delay. Mechatronics, 20(7), 812– 823. Secchi, C., Straimgioli, S., and Fantuzzi, C. (2005). Power scaling in port-hamiltonian based telemanipulation. In 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems, 1850–1855. IEEE. Speich, J.E. and Goldfarb, M. (2002). Implementation of loop-shaping compensators to increase the transparency bandwidth of a scaled telemanipulation system. In Proceedings 2002 IEEE International Conference on Robotics and Automation, volume 3, 2886–2893. IEEE. Vander Poorten, E.B., Yokokohji, Y., and Yoshikawa, T. (2006). Stability analysis and robust control for fixedscale teleoperation. Advanced Robotics, 20(6), 681–706. Yan, J. and Salcudean, S.E. (1996). Teleoperation controller design using h-infinity optimization with application to motion-scaling. IEEE Transactions on control systems technology, 4(3), 244–258.
the energy plot E2port confirms the passivity of the time domain scaling controller. Figure 11 and Fig. 12 present the energy-based Approach 2 with velocity (6) and force scaling adaptation (7) respectively. The passivity of the scaling is confirmed by the solid ∗ energy plot E2port . The dashed energy plot ESt presents the reference energy which is reset when the positions of master and slave match. The master leads in Fig. 11 at t = [0.5s − 2.25s] and in Fig. 12 at t = [2.1s − 4.6s]. The positive aspect of Approach 2 is that the scaling does not change during the wall contact in contrast to the powerbased approaches. The experiment in Fig. 13 presents a delayed setup with TDPA at 100ms roundtrip-delay with power-based scaling (Approach 1) and velocity scaling adaptation (4). A pure motion scaling (ν des = 0.7, µdes = 1) has been applied. Since E4L2R ≤ E2L2R (compare Fig. 7) the energy plot confirms that the admittance type PC ensures a passive communication. The scaling ν ∗ acts on the velocity output of the PC. The resulting velocity scaling ν is ν P C when ν P C < ν des and ν des when ν P C > ν des . When power flows from slave to master in R2L direction, the velocity scaling is deactivated (ν = 1). The consideration of the PC dissipation in the motion scaling leads to lower conservatism. 5. CONCLUSION A new time domain passivity control for scaling subsystems has been proposed that focuses on setups that require a down-scaling of motions for higher precision or a smaller workspace of the slave robot. A power- and an energy-based control approach have been presented that can be applied for impedance and pure motion scaling with adaptation of force or velocity. The user has to decide for a force or velocity adaptation according to the respective teleoperation application. A time delay control approach has been extended with the proposed scaling method. The time domain control approach has been successfully validated in experiments. In contrast to former approaches, in the proposed method, a variable scaling is feasible, which is otherwise very difficult to integrate in an alternative frequency based stability analysis. Furthermore, no passivity assumptions for environments have to be made. REFERENCES Boukhnifer, M. and Ferreira, A. (2006). Stability and transparency for scaled teleoperation system. In 2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, 4217–4222. IEEE. Boukhnifer, M., Ferreira, A., and Fontaine, J.G. (2004). Scaled teleoperation controller design for micromanipulation over internet. In IEEE International Conference on Robotics and Automation, 2004. Proceedings. ICRA’04. 2004, volume 5, 4577–4583. IEEE. Colgate, J.E. (1991). Power and impedance scaling in bilateral manipulation. In Proceedings. 1991 IEEE International Conference on Robotics and Automation, 2292–2297. IEEE. Goldfarb, M. (1998). Dimensional analysis and selective distortion in scaled bilateral telemanipulation. In 60
ScienceDirect
IFAC PapersOnLine 52-18 (2019) 31–36
Time Domain Control for Passive Variable Time Time Domain Domain Control Control for for Passive Passive Variable Variable Motion and Force Scaling in Delayed Time Domain Control for Passive Variable Motion and Force Scaling in Delayed Motion and Force Scaling in Delayed Motion andTeleoperation Force Scaling in Delayed Teleoperation Teleoperation Teleoperation M. Panzirsch ∗∗ R. Balachandran ∗∗
M. Panzirsch ∗∗ R. Balachandran ∗∗ M. M. Panzirsch Panzirsch ∗ R. R. Balachandran Balachandran ∗ M. Panzirsch R. BalachandranAerospace Robotics and Mechatronics, Robotics and Mechatronics, German German Aerospace Center Center Robotics and and Mechatronics, German Aerospace Center Center Wessling, Germany (e-mail: [email protected]). Robotics Mechatronics, German Aerospace Wessling, Germany (e-mail: [email protected]). Institute of Wessling, Robotics and Mechatronics, Aerospace Center (DLR), 82234 Wessling, Germany (e-mail: German [email protected]). (DLR), 82234 Germany (e-mail: [email protected]). (DLR), 82234 Wessling, Germany (e-mail: [email protected]). Abstract: Abstract: Scaling Scaling of of motion motion and and forces forces has has always always been been of of high high relevance relevance in in teleoperation teleoperation setups setups Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Abstract: Scaling of motion and forces has always been of high relevance in teleoperation setups since it allows the adaptation of workspaces of master and slave devices or to increase precision. Teleoperation setups are often affected by a delay in the communication channel. Most state since it allows the adaptation of workspaces of master and slave devices or to increase precision. Teleoperation setups are oftenof affected by aofdelay in and the slave communication channel. Most state since itart allows the adaptation workspaces master devices orbased to increase precision. Teleoperation setups are often oftenthat affected by aa stability delay in despite the communication communication channel. Most state of the control approaches guarantee delay are on the passivity Teleoperation setups are affected by delay in the channel. Most state of the art control approaches that guarantee stability despite delay are based on the passivity Teleoperation setups are often affected by a delay in the communication channel. Most state of the the art artwhich control approaches that guarantee guarantee stability despite delay arepaper basedproposes on the the passivity passivity criterion is highly restrictive to standard scaling methods. This different of control approaches that stability despite delay are based on criterion which is highly restrictive to standard scalingdespite methods. This paper proposes different of the art control approaches that guarantee stability delay are based on the passivity criterion which is highly restrictive to standard scaling methods. This paper proposes different time domain control concepts that the force the flow criterion which is highly restrictive to standard scalingor methods. This based paper on proposes different time domain control concepts that regulate regulate the motion motion force scaling scaling based the energy energy flow criterion which is highly restrictive toapproach standard scalingor methods. This motion paper on proposes different time domain control concepts that regulate the or force scaling based on the flow in delayed teleoperation systems. focuses on setups with down-scaling and time domain control concepts thatThe regulate the motion motion or force scaling based on the energy energy flow in delayed teleoperation systems. The approach focuses on setups with motion down-scaling and time domain control concepts that regulate the motion or force scaling based on the energy flow in applicable delayed teleoperation teleoperation systems. The impedance approach focuses focuses onThe setups with control motion is down-scaling anda is to variable motion and scaling. scaling integrated in in delayed systems. The approach on setups with motion down-scaling and is applicable to variable motion and impedance scaling. The scaling control is integrated in in delayed teleoperation systems. The approach focuses onThe setups with motion down-scaling andaaa is applicable applicable to time variable motion and impedance scaling. The scaling control is integrated in in state of the art delay control concept and its performance is analyzed in experiments. is to variable motion and impedance scaling. scaling control is integrated state of the art time delay control concept and its performance is analyzed in experiments. is applicable to time variable motion impedance The scaling controlin integrated in a state of delay control concept and performance is state of the the art art time delay controland concept and its itsscaling. performance is analyzed analyzed inisexperiments. experiments. Copyright 2019. Thedelay Authors. Published by Elsevier All rights reserved. state of the© art time control concept and itsLtd. performance is analyzed in experiments. Keywords: Keywords: Scaling, Scaling, Time Time Delay, Delay, Teleoperation, Teleoperation, Time Time Domain Domain Control Control Keywords: Keywords: Scaling, Scaling, Time Time Delay, Delay, Teleoperation, Teleoperation, Time Time Domain Domain Control Control Keywords: Scaling, Time Delay, Teleoperation, Time Domain Control 1. INTRODUCTION A large set of 1. INTRODUCTION A large set of control control concepts concepts for for delayed delayed teleoperation teleoperation 1. A based large set set of control concepts for delayed teleoperation is on an energy criterion. Colgate (1991) and 1. INTRODUCTION INTRODUCTION A large of control concepts for delayed teleoperation is based on an energy criterion. Colgate (1991) and Itoh Itoh 1. INTRODUCTION A large set control concepts for delayed teleoperation is based based on of anchose energy criterion. based Colgate (1991) and Itoh Itoh et al. (2000) a passivity design considering is on an energy criterion. Colgate (1991) and et al. (2000) chose a passivity based design considering is based on an energy criterion. Colgate (1991) and Itoh et al. (2000) chose aaInpassivity based design considering aa scattering matrix. the system of Itoh et al. (2000), et al. (2000) chose passivity based design considering Teleoperation is a mature technology that has a large scattering matrix. Inpassivity the system of design Itoh et considering al. (2000), al. (2000) chose a based Teleoperation is a mature technology that has a large et a scattering matrix. In the system of Itoh et al. (2000), the scaling design was specific in that it was separated a scattering matrix.was In specific the system of Itoh et al. (2000), Teleoperation is technology has aa large variety of application application fields ranging ranging fromthat nuclear research scaling design in that it was separated Teleoperation is aa mature mature technology that has research large the matrix.was In specific the system of Itoh etmethod al. (2000), variety of fields from nuclear thescattering scaling design was specific in that that it was was separated Teleoperation is medical a mature technology that has a large athe from the telemanipulator. The wave variables was scaling design in it separated variety of application fields ranging from nuclear research to space and the sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was variety of application fields ranging from nuclear research the scaling design was specific in that it was separated to space and the medical sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was variety of application fields ranging from nuclear research applied by Boukhnifer and Ferreira (2006). Secchi et al. from the telemanipulator. The wave variables method was to space and the medical sector. Some prefer an up-scaling of the master motions for a large slave robot workspace, applied by Boukhnifer and Ferreira (2006). Secchi et al. to space and the medical sector. Some prefer an up-scaling from the telemanipulator. The wave variables method was of the master motions forsector. a large slave robot workspace, applied by Boukhnifer and Ferreira (2006). Secchi et al. to space and the medical Some prefer an up-scaling (2005) designed a scaled teleoperation setup with the portapplied by Boukhnifer and Ferreira (2006). Secchi et al. of the master motions for a large slave robot workspace, others as surgical applications require microor nano(2005) designed a scaledand teleoperation setup with the et portof the master motions for a largerequire slave robot workspace, applied by Boukhnifer Ferreira (2006). Secchi al. others as surgical applications microor nano(2005) designed a scaled teleoperation setup with the portof the master motions for a large slave robot workspace, Hamiltonian system representation. Often, it is assumed (2005) designed a scaled teleoperationOften, setup with the portothers as surgical applications require microor nanomanipulation capabilities in the extreme case (Onal and Hamiltonian system representation. it is assumed others as surgical applications require microor nano(2005) designed a scaled teleoperation setup with the portmanipulation capabilities in the extreme case (Onal and Hamiltonian systemare representation. Often, it is is assumed assumed others as surgical applications or downnanothat environments passive et Secchi system representation. it manipulation capabilities the extreme case (Onal and Sitti (2009)) which can be bein achieved by micromotion that environments are passive (Itoh (Itoh Often, et al. al. (2000), (2000), Secchi manipulation capabilities in achieved the require extreme (Onal and Hamiltonian Hamiltonian system representation. Often, it is assumed Sitti (2009)) which can by aacase motion downthat environments are passive (Itoh et al. (2000), Secchi manipulation capabilities in the extreme case (Onal and et al. (2005), Onal and Sitti (2009)) which presents a clear that environments are passive (Itoh et al. (2000), Secchi Sitti (2009)) which can be achieved by a motion downscaling. Apart from motion scaling also force scaling can be et al. (2005), Onal and Sitti (2009)) which presents a clear Sitti (2009)) which can be achieved by a motion downthat environments are Sitti passive (Itoh et al.For (2000), Secchi scaling. Apartwhich from motion scaling alsoby force scaling can be et al. (2005), Onal and (2009)) which presents a clear Sitti (2009)) can be achieved a motion downlimitation for the teleoperation scenario. example, a et al. (2005), Onal and Sitti (2009)) which presents a clear scaling. Apart from motion scaling also force scaling can be helpful, for example, in teleoperation training or for task limitation forOnal the teleoperation scenario. For example, a scaling. Apart from motion scaling alsotraining force scaling can be et al. (2005), and Sitti (2009)) which presents a clear helpful, for example, in teleoperation or for task limitation for the teleoperation scenario. For example, a scaling. Apart from motion scaling also force scaling can be beating heart in a medical scenario or a human interacting limitation for in the teleoperation scenario. For example, a helpful, for example, in teleoperation training or for task allocation in multilateral setups (Panzirsch et al. (2018)). beating heart a medical scenario or a human interacting helpful, for example, in teleoperation training or for task limitation for the teleoperation scenario. For example, a allocation in multilateral setups (Panzirsch et al. (2018)). beating heart in a medical scenario or a human interacting helpful, for example, in teleoperation training or for task with a robot in an ambient assisted living scenario presents beating heart in a medical scenario or a human interacting allocation in multilateral setups (Panzirsch et al. (2018)). Depending on the combination of motion and force scaling with a robot in an ambient assisted living scenario presents allocation in multilateral setups (Panzirsch et al. (2018)). beating heart in a medical scenario or a human interacting Depending on the combination of motion and force scaling with a robot in an ambient assisted living scenario presents allocation inon setupsof (Panzirsch etforce al. (2018)). an active environment. Also, several control approaches a robot in an ambient assisted living scenario presents Depending the combination motion scaling factors, so-called power scaling, impedance scaling or pure with an active environment. Also, several control approaches Depending onmultilateral thepower combination ofimpedance motion and and force scaling with a robot in an ambient assisted living scenario presents factors, so-called scaling, scaling or pure an active environment. Also, several control approaches Depending on the combination of motion and force scaling require power scaling settings (Boukhnifer and Ferreira an active environment. Also, several control approaches factors, so-called power scaling, impedance scaling or pure motion or force scaling can be designed. The works of require power scaling settings (Boukhnifer and Ferreira factors, so-called or pure active environment. Also, several control approaches motion or force power scalingscaling, can beimpedance designed. scaling The works of an require power scaling settings (Boukhnifer and Ferreira factors, so-called power scaling, impedance scaling or pure (2006)) since it is an intrinsically passive functionality. But require power scaling settings (Boukhnifer and Ferreira motion force scaling can works Colgate (1991), Vander Poorten etdesigned. al. (2006) (2006)The and Goldfarb Goldfarb since it scaling is an intrinsically passive functionality. But motion or or forceVander scalingPoorten can be beet designed. The works of of (2006)) require power settings (Boukhnifer and Ferreira Colgate (1991), al. and (2006)) since it is an intrinsically passive functionality. But motion or force scaling can be designed. The works of the power scaling values differ extremely from impedance (2006)) since it is an intrinsically passive functionality. But Colgate (1991), Vander Poorten et al. (2006) and Goldfarb (1998) discuss different force scaling designs according to the power scaling values differ extremely from impedance Colgate (1991), Vander Poorten et al. (2006) and Goldfarb (2006)) since itwhich is an intrinsically passive functionality. But (1998) discuss different force scaling designs according to the power scaling values differ extremely from impedance Colgate (1991), Vander Poorten et al. (2006) and Goldfarb scaling values are often preferable. the power scaling values differ extremely from impedance (1998) discuss different force scaling designs according to the properties of the the environmental environmental impedance that can can values which are often preferable. (1998) discuss different force scaling designs according to scaling the power scaling values differ extremely from impedance the properties of impedance that scaling values which are often preferable. (1998) discuss different force scaling designs according to values which are often preferable. the properties of environmental that be primarily inertial, or exhibit exhibitimpedance viscous damping. damping. theprimarily propertiesinertial, of the the elastic environmental impedance that can can scaling Here, focus on passivity criterion values often preferable. be or viscous Here, we we focuswhich on the theare passivity criterion since since most most state state the propertiesinertial, of the elastic environmental impedance that can scaling be primarily elastic or exhibit viscous damping. be primarily inertial, elastic or exhibit viscous damping. Here, we focus on the passivity criterion since most of the art approaches for time-delayed teleoperation as Here, we focus on the passivity criterion since most state state Scaling factors have been applied within different teleopof the art approaches for time-delayed teleoperation as be primarily inertial, elastic or exhibit viscous damping. Here, we focus on the passivity criterion since or most state Scaling factors have been applied within different teleop- the of the art approaches for time-delayed teleoperation as wave variables method (Niemeyer (1996)) the time of the art approaches for time-delayed teleoperation Scaling factors have been applied within different teleoperation control architectures and with a variety of stability the wave variables method (Niemeyer (1996)) or the time Scaling control factors architectures have been applied within different teleop- of the art approaches for time-delayed teleoperation as as eration and with a variety of stability the wave variables method (Niemeyer (1996)) or the time Scaling factors have been applied within different teleopdomain passivity approach (TDPA, Ryu et al. (2010), the wave variables method (Niemeyer (1996)) or the time eration architectures and variety of stability stability domain passivity approach (TDPA, Ryu et al. (2010), analyzes. In Speich Speich and Goldfarb Goldfarb (2002), the transparency transparency eration control control architectures and with with aa variety of the wave variables method (Niemeyer (1996)) or the time analyzes. In and (2002), the domain passivity passivity approach (TDPA, Ryu et al. al. (2010), (2010), eration control architectures and with a with variety of stability domain Panzirsch et are based this The approach et analyzes. In Speich and (2002), the transparency of scaled position-force architecture loop-shaping Panzirsch et al. al. (2019)) (2019)) are (TDPA, based on on Ryu this criterion. criterion. The analyzes. In position-force Speich and Goldfarb Goldfarb (2002), the transparency domain passivity approach (TDPA, Ryu et al.concepts (2010), of aa scaled architecture with loop-shaping Panzirsch et al. (2019)) are based on this criterion. The analyzes. In Speich and Goldfarb (2002), the transparency benefit is the high modularity of energy-based Panzirsch et al. (2019)) are based on this criterion. The of a scaled position-force architecture with loop-shaping compensators has been evaluated. Also, the H approach benefit is the high modularity of energy-based concepts of a scaled position-force architecture with loop-shaping ∞ Panzirsch et al. (2019)) are based on this criterion. The compensators has been evaluated. Also,with the Hloop-shaping ∞ approach benefit is the high modularity of energy-based concepts of a scaled position-force architecture which allows, for example, uncomplicated extensions of benefit is the high modularity of energy-based concepts compensators has been evaluated. Also, the H approach (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and ∞ which allows, for example, uncomplicated extensions of approach compensators has been evaluated. Also, the H ∞ benefit is the high modularity of energy-based concepts (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and which allows, for example, uncomplicated extensions of approach compensators has been evaluated. Also, the H bilateral to multilateral setups (Panzirsch et al. (2013)). which allows, for example, uncomplicated extensions of ∞ (Yan and Salcudean Salcudean (1996), Boukhnifer et al. al. (2004)) (2004)) and bilateral to multilateral setups (Panzirsch et al. (2013)). a sliding mode control (Khan et al. (2009)) have been (Yan and (1996), Boukhnifer et and which allows, for propose example, uncomplicated extensions of a sliding mode control (Khan et al. (2009)) have been bilateral to multilateral setups (Panzirsch et al. (2013)). (Yan and Salcudean (1996), Boukhnifer et al. (2004)) and In this paper, we a time domain control concept bilateral to multilateral setups (Panzirsch et al. (2013)). a sliding mode control (Khan et al. (2009)) have been implemented for scaled teleoperation. Power scaling was In this paper, we proposesetups a time(Panzirsch domain control concept a sliding mode control (Khan et al. (2009)) have been bilateral to multilateral et al. (2013)). implemented for scaled teleoperation. Power scaling was In this paper, we propose a time domain control concept a slidingbymode (Khan et al. (2009)) have been for passive scaling that be teleoperation paper, we propose a time domain in control concept implemented for scaled teleoperation. Power was applied Ferreira Boukhnifer for this passive scaling that can can be applied applied in teleoperation implemented forcontrol scaled and teleoperation. Power scaling scaling was In paper, we propose a time domain control concept applied by Boukhnifer Boukhnifer and Ferreira (2006), (2006), Boukhnifer for passive scaling that can be in teleoperation implemented for scaled teleoperation. Power scaling was In scenarios with active environments. In contrast to former for this passive scaling that can be applied applied in teleoperation applied by Boukhnifer Boukhnifer and Ferreira (2006), Boukhnifer et al. (2004) and Jazayeri and Tavakoli (2013). Jazayeri scenarios with active environments. In contrast to former applied by and Ferreira (2006), Boukhnifer for passive scaling that can be applied in teleoperation et al. (2004) and Jazayeri and Tavakoli (2013). Jazayeri scenarios with active environments. In contrast to former applied by Boukhnifer and Ferreira (2006), Boukhnifer approaches, the concept guarantees passivity in setups scenarios with active environments. In contrast to former et al. al.Tavakoli (2004) (2013) and Jazayeri Jazayeri and Tavakolistability (2013).for Jazayeri and focused on absolute scaled approaches, the concept guarantees passivity in setups et (2004) and and Tavakoli (2013). Jazayeri with active environments. Inpassivity contrast to former and Tavakoli (2013) focused and on absolute stability for scaled scenarios approaches, the concept guarantees passivity in setups et al. (2004) and Jazayeri Tavakoli (2013). Jazayeri with pure motion, pure force and impedance scaling, and approaches, the concept guarantees in setups and Tavakoli (2013) focused on absolute stability for scaled sampled-data systems. Impedance scaling has been conwith pure motion, pure force and impedance scaling, and and Tavakoli (2013) focused on absolute stability for scaled approaches, the concept guarantees passivity in present setups sampled-data systems. Impedance scaling has been con- with pure motion, pure force and impedance scaling, and and Tavakoli (2013) focused on absolute stability for scaled allows for time-varying scaling designs. Also, we with pure motion, pure force and impedance scaling, and sampled-data systems. Impedance scaling has been considered by Onal and Sitti (2009). Vander Poorten et al. allowspure for motion, time-varying scaling designs. Also,scaling, we present sampled-data systems. Impedance scaling Poorten has beenetconwith pure force and impedance and sidered by Onal and Sitti (2009). Vander al. allows for time-varying scaling designs. Also, we present sampled-data systems. Impedance scaling has been conhow the conservatism of time domain passivity control forconservatism time-varying ofscaling designs. Also, we present sidered performed by Onal Onal and and Sitti (2009). (2009). Vander Poorten et on al. allows (2006) an absolute stability analysis based how the time domain passivity control sidered by Sitti Vander Poorten et al. allows for time-varying scaling designs. Also, we present (2006) performed an absolute stability analysis based on howdelayed the conservatism conservatism ofmotion time domain domain passivity control sidered by Onal and Sitti (2009). Vander Poorten et on al. how in setups scaling can the time control (2006) performed an absolute absolute stability analysis based on the scattering matrix of fixed-scale teleoperin delayed setups with withof scaling passivity can be be reduced. reduced. (2006) performed an stability analysis based how the conservatism ofmotion time domain passivity control the scattering matrix of a a delay-free, delay-free, fixed-scale teleoperin delayed setups with motion scaling can be reduced. (2006) performed an absolute stability analysis based on in delayed setups with motion scaling can be reduced. the scattering matrix of a delay-free, fixed-scale teleoperation setup. the of a delay-free, fixed-scale teleoper- in delayed setups with motion scaling can be reduced. ationscattering setup. matrix the scattering matrix of a delay-free, fixed-scale teleoperation ation setup. setup. ation setup. 2405-8963 Copyright © 2019. The Authors. Published by Elsevier Ltd. All rights reserved. ∗ ∗ Institute of Institute of ∗ ∗ Institute of (DLR), 82234 of ∗ Institute (DLR), 82234
Copyright © 2019 IFAC 55 Copyright © under 2019 IFAC 55 Control. Peer review responsibility of International Federation of Automatic Copyright © 55 Copyright © 2019 2019 IFAC IFAC 55 10.1016/j.ifacol.2019.12.202 Copyright © 2019 IFAC 55
2019 IFAC TDS 32 Sinaia, Romania, September 9-11, 2019
Scaling ν
vm
ν vm + -
Human M aster Fh
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
vs
Ctrl
+ Fcµ
Slave
Env.
+ -
µ
Fe
Fc
Fig. 1. Signal flow diagram of a 2-Channel architecture with scaling. Agent Λ1
Fig. 3. Experimental setup: two 1-DoF rotational devices.
Agent Λ2
v1
v1
v2
v3
v3
+
+
+
+
+
F2
Ctrl F3
+ -
Human Fh M aster F1 l P ort 1
Scaling
-
-
P ort 2
P ort 3
Slave
Fe
device (Zm = µFc vm ). With vm = νvs , impedance scaling requires µ = 1/ν which is highly contradictive to the passive power scaling. The experiment in Fig. 5 presents an impedance scaling teleoperation with ν = 0.5 and µ = 2. The energy plot clearly shows the energy generation during the operation (negative slope of E2port ) when the slave leads the motion at t > 3.5s.
Env.
-
The impedance scaling focuses on the haptic perception of impedance at the master device and neglects the visual feedback. Alternatively, it can be argued that the human operator relates the force displayed at the master device to the motion of the slave robot (instead of the master motion). Then, a force scaling of µ = 1 is reasonable. This may be especially relevant for master and slave systems with similar workspace where a pure motion scaling is applied to increase the precision of the operation. In teleoperation scenarios with free motion and soft walls, the pure motion scaling promises a good perception of the environmental interaction whereas during hard contacts (low slave motion) the impedance scaling might be more intuitive. The reader is referred to the work of Goldfarb (1998) for more details on the choice of force scaling factors depending on the environment contact. As visible from the experiment in Fig. 6 and the energy plot E2port , the pure motion scaling (µ = ν, µ = 1, ν = 0.5) is not passive.
Fig. 2. Network representation of a 2-Channel architecture with scaling.
The method promises high transparency especially during motion down-scaling. The paper is structured as follows: Section 2 analyzes the energetic behavior of motion and force scalings in control networks. A power-based and an energy-based time domain passivity control for variable motion and force scaling are introduced in Section 3. The experimental evaluation which includes time delay teleoperation setups is presented in Section 4. Finally, Section 5 concludes the work. 2. PROBLEM STATEMENT The signal flow diagram of a scaled teleoperation setup and the respective network representation are depicted in Fig. 1 and Fig. 2 respectively. A Human uses a M aster input device to control a slave robot in its environment. The two devices can be coupled with a PD controller (Ctrl). The variables ν and µ scale the motion (velocity) and force respectively (ν, µ ∈ [0, ∞[). The following relations ν between the signals holds: v1 = vm , v2 = vm , v3 = v s , µ F1 = Fc , F2 = F3 = Fc . The network representation is an electrical diagram that allows the power observation at the ports i with power-conjugated signals force Fi and velocity vi . The Scaling as well as the Ctrl are 2-port networks.
Depending on the applications, also a variable velocity scaling design can be reasonable. For example, the velocity scaling can be a function of the master velocity (ν = f (vm )) such that for fast master motion ν = 1 since it can be assumed that the human wants to move far. Whereas at lower master motion, the velocity scaling can be faded down to ν = 0.5 since the human wants to command a precise motion.
In the following, a passivity control for arbitrary and variable scaling values is designed that leads to a passive scaling network subsystem that can be applied to passivity-based control frameworks in a highly modular manner.
Concerning an analytical analysis of a scaling network in the frequency domain (e.g. Raisbeck’s criterion), the scaling subsystem is only passive if the motion and force scaling values are equal (power scaling, µ = ν). That means that, for the sake of passivity, in case of motion up-scaling, the force feedback has to be scaled up and in case of motion down-scaling, the feedback force has to be scaled down. The experiment in Fig. 4 shows that with equal µ and ν (µ = ν = 0.5), the energy over the scaling 2-port E2port is always zero which confirms its passivity. The experiments were performed with two 1-DoF rotational devices (SENSODRIVE, compare Fig. 3). The coupling software was executed on a QNX host machine at a sampling rate of 1kHz.
3. PROPOSED TIME DOMAIN CONTROL APPROACH As it can be analyzed from the energy plots in Fig. 4 to Fig. 6, the scaling dissipates or generates (positive and negative slope of E2port respectively) energy depending on the scaling values and on which device leads the motion. The power dissipation and generation can be calculated as follows (if ν, µ = 0): ν < P1 , if ν < µ (case 1) (1) P2 = P1 ≥ P1 , if ν ≥ µ (case 2). µ
With the classical impedance scaling, the slave side impedance (Zs = Fc vs ) can be displayed at the master 56
−2 1
1.5
2
2.5
3
3.5
4
4.5
−0.5 −1 −1.5 1
1.5
2
2.5
3
3.5
4
4.5
5
1
3
4
5
−1 −1.5 1
2
3
4
5
6
1.5
2
2.5
3
3.5
4
4.5
µForce νVelocity
1 0.5 0 0.5
1
1.5
2
2.5
3
3.5
4
4.5
0 2
4
5
1.5
2
2.5
3
3.5
4
4.5
0.5
1
1.5
2
2.5
3
3.5
4
4.5
0.5
1
1.5
2
2.5
3
3.5
4
4.5
−1
0
6
µForce νVelocity
1 0 2
3
4
Time [s]
Time [s]
Fig. 4. Power scaling.
3
2
1
5
1
−0.5 1
5
0.5
−1.5
Scaling [−]
1
−2
−0.5
E2port [J ]
0.2
Scaling [−]
0.5
Master Slave
0
0.5
0.4
0
2
0.5 0
0.6
0.1
33
6
−0.5
E2port [J ]
E2port [J ]
2
0.5 0
0.2
Scaling [−]
Master Slave
−2
5
0.5 0
0.5
0
P osition [rad]
Master Slave
2
Torque [Nm]
0 −1
P osition [rad]
1
0.5
Torque [Nm]
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Torque [Nm]
P osition [rad]
2019 IFAC TDS Sinaia, Romania, September 9-11, 2019
Fig. 5. Impedance scaling.
5
6
1
µForce νVelocity
0.5 0 0.5
1
1.5
2
2.5
3
3.5
4
4.5
Time [s]
Fig. 6. Pure motion scaling.
that this adaptation is not problematic: In this work, we focus on a down-scaling of the velocity such that the teleoperated slave robot moves slower than the master device. Thus, the controller has to adapt (increase) the velocity scaling if energy is generated by the scaling. As analyzed before, the scaling generates energy if energy is flowing from slave to master, that means if the master device moves out of a wall contact or if the slave leads the motion. If the scaling is not varied (time domain control inactive), the master moves much faster than the slave, which might be dangerous if the slave leads the motion. In contrast, if the velocity scaling is increased in such situations (by the passivity control action), the master device moves as slow and as far as the slave which is a much safer procedure.
In case 1, energy is dissipated in the left to right energy flow direction (L2R) and generated in the right to left (R2L) energy flow direction. In case 2, it behaves vice versa. The power flow direction can be distinguished by the sign of the power Pi . The power values PiL2R and PiR2L are positive by definition: Pi , if Pi > 0 L2R Pi = (2) 0 , if Pi ≤ 0, −Pi , if Pi < 0 (3) PiR2L = 0 , if Pi ≥ 0. In typical teleoperation setups, the master moves the slave such that in case 1 (for example at pure motion downscaling) the scaling is mostly overall passive. Still, classical frequency-based stability approaches do not consider this power flow direction dependency and therefore a passive power scaling (ν = µ) has to be chosen or active environments are not allowed. Applying a time domain energy observation and control, this conservatism can be heavily reduced. In the following, different time domain control approaches for case 1 in (1) will be discussed that consider different power and energy criteria. Furthermore, they differ by the adaptation of motion or force scaling. Finally, the integration of the control approach in a state of the art time delay control approach is presented.
Now, two approaches that consider a power-based and an energy-based criterion respectively are presented. Power-based passivity criterion (Approach 1): As explained before, the power flow direction can be analyzed and the focused motion down-scaling produces energy only in the R2L flow direction, which means when the slave leads the master. Therefore, a passivity controller can be designed that adapts the scaling for the sake of passivity to passive power scaling (µ = ν) if power flows in R2L direction: The velocity scaling des ν , if P1L2R > 0 ν(k) = (4) µ , if P1L2R = 0, or alternatively, the force scaling can be adapted, therefore: des , if P1L2R > 0 µ (5) µ(k) = ν , if P1L2R = 0. Note that force scaling may lead to disturbances in the teleoperation setup and therefore, velocity scaling should be favoured. Still, for the sake of completeness, the force scaling is also presented here. Experiments concerning impedance and pure motion scaling are presented in Section 4.
3.1 Time domain control for passive scaling In the following section, we concentrate on case 1 (ν < µ) settings. The proposed control method preserves passivity through the online adaptation of the velocity or force scaling as soon as passivity is violated. Therefore, considering motion down-scaling, the controller has to increase the velocity scaling or to decrease the force scaling in phases of undesired energy generation to achieve ν = µ (passive power scaling). In general, the force scaling should remain constant during teleoperation since the forces are directly perceived at the master device. In contrast, the velocity adaptation is not as obvious and disturbing to the operator since the operator focuses on the position command which is the integral of the adapted variable. On the first sight, the up-scaling of the velocity by the controller may appear as an unintuitive solution. But, a closer analysis reveals
The negative aspect of the power-based controller is that the scaling is not only varied when the slave leads the motion but also when the master leaves a wall contact (P1L2R = 0). Note that if the ν adaptation is chosen, 57
2019 IFAC TDS 34 Sinaia, Romania, September 9-11, 2019
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Agent Λ1
Agent Λ2
v1
v1
v2
v3
v4
v5
v6
v6
+
+
+
+
+
+
+
+
F3 P C2 F4 Scaling F5
Ctrl F6
CC
+ -
Human Fh M aster F1 P C1 F2 l P ort 1
P ort 2
-
-
P ort 3
P ort 4
P ort 5
Slave
-
Fe
Env.
-
P ort 6
Fig. 7. Network representation of the TDPA (Ryu et al. (2010)) for delayed teleoperation with scaling. the slave leaves the wall contact faster. As soon as the wall contact is over (P1L2R > 0), the standard scaling is reactivated. For example, in pick and place tasks, the operator wants to feel a wall contact, but also wants to leave it as fast as possible. Then, the power-based velocity adaptation has no negative effect. In soft environments, on the other hand, the slave leaves the wall contact faster and the perception of the wall contact is altered. Still, this may not be critical since the impedance of an object is mainly analyzed in the penetration direction.
time domain passivity control (TDPA, Ryu et al. (2010)). The TDPA introduces passivity observers that measure the energy that has been generated by the CC and passivity controllers (PC) that dissipate energy to ensure the passivity of the two-port between ports 1 and 4. This approach also considers two directions of energy flow and therefore, two PCs are implemented. An impedance-type P C1 varies the force which is sent to the master side with a variable damping α that depends on the energy that has to be dissipated and on the velocity v2 at the P C1 port. An PC admittance-type P C2 varies the velocity (v4 = v3 − vm ) which is sent to the slave side with a variable damping β that is calculated from the energy that has to be dissipated and the force F3 at the P C2 port. Since this variation can be interpreted as a down-scaling ν P C
Energy-based passivity criteria (Approach 2): This negative aspect of the power-based controller motivates for another solution that circumvents the scaling adaption during one wall contact. In contrast to the power-based design (Approach 1), the following concept considers the potential energy storage in the coupling controller that results from the coupling controller’s spring-like element.
PC del ν P C = min(vm /vm , 1),
(calculated from the delayed master reference velocity del PC vm = v3 and the velocity output vm of the P C2), it can be considered for the desired velocity scaling ν ∗ that PC acts on vm in case of velocity scaling adaptation (4) or (6): 1 , if ν P C < ν des ∗ ν = (9) des PC min(1, ν /ν ) , if ν P C ≥ ν des ,
This energy storage Est is built up, for example, during a wall contact: Est (k) = Est (k − 1) + (P2L2R (k) + P3R2L (k) − P3L2R (k) − P2R2L (k))Ts , with the sampling time Ts (compare Fig. 2).
During a wall contact, energy is charged into the coupling controller’s spring. The same amount of energy is released when the penetration into the wall is reduced. Therefore, a time domain controller for passive scaling that considers the energy ESt instead of power does not vary the scaling during the wall contact (in contrast to Approach 1). The velocity scaling
such that the effective overall scaling ν ∗ ν P C of the dedel layed master velocity vm is not higher than ν des (if the time domain controller for passive scaling is not active, otherwise ν = µ). The consideration of the P C2 scaling in the desired motion scaling ν leads to a less conservative passivity control since the P C2 dissipation leads to a position drift (p4 = p3 ) after integration of v4 . If the PC dissipation would not be considered in ν = ν ∗ ν P C , the velocity would be further reduced and position drift increased. The consideration of a force scaling adaptation (5) or (7) in P C1 is not presented here.
ν des , if Est (k) ≥ 0 (6) µ , if Est (k) < 0, or alternatively, the force scaling can be adapted therefore: µdes , if Est (k) ≥ 0 (7) µ(k) = ν , if Est (k) < 0. ν(k) =
(8)
4. EXPERIMENTAL EVALUATION
To assure that the scaling switches when the leading role switches from master to slave or vice versa, the reference ∗ has to be reset when xm = xs (with a energy storage ESt certain threshold). If Est (k) = 0, the old values of ν and µ have to be set. In contrast to the power-based adaption (Approach 1), force scaling may lead to lesser disturbances in Approach 2 since the scaling is not adapted during a wall contact.
The first experiment (see Fig. 8) presents the power scaling method (Approach 1) with pure motion scaling (ν des = 0.5, µdes = 1) and velocity scaling adaptation (4). The master moves the slave device in free motion and against a wall (t = [2s − 6s]). The velocity scaling is adapted when the master leaves the wall penetration, since power is flowing from slave to master (P1L2R = 0). Afterwards (t = [6s − 7.5s]), the slave moves the master and the velocity scaling is set to 1 (P1L2R = 0). With this control strategy, a passive scaling can be achieved, as it is confirmed by the purely positive 2-port energy E2port of the scaling subsystem.
3.2 Scaling in Time Delay Setups Figure 7 presents the network representation of a delayed teleoperation setup with communication channel CC and 58
5
4
6
5
7
6
7
3
4
5
6
7
8
0.2 0.1 0 2
3
4
5
7 Force Velocity
6
0.5 0 2
3
4
5
6
7
8
2
2.5
3
3.5
4
4.5
5
5.5
0 −1 1.5 1.5
2
2.5
3
3.5
4
4.5
5
1
0 1.5 1
2
2.5
3
3.5
4
4.5
5.5
5
5.5
0.5 0 1.5 2
2
1
2.5
3
3.5
4
4.5
5
2
2.5
3
Time [s]
3.5
4
4.5
5
Master Slave
0 1
5.5
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
1
2
3
4
5
6
3
4
5
6
0
L2R R2L
5
0 0 0.1
0.05 0 0
5.5
µForce νVelocity
0 1.5
2 1 0 −1
−1 0
L2R R2L
0.5
8
1
P osition [rad]
P osition [rad] 8
L2R R2L
0.5 0 2
8
35
Power [J /s] Torque [Nm]
3
4
Master Slave
E2port [J ]
2 1
3
3 2 1 0 −1 1.5 1
Scaling [−]
Scaling [−]
E2port [J ] Power [J /s] Torque [Nm]
2 0.2 0 −0.2 −0.4
Master Slave
E2port [J ] Power [J /s] Torque [Nm]
2 1 0 −1
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Scaling [−]
P osition [rad]
2019 IFAC TDS Sinaia, Romania, September 9-11, 2019
1
µForce νVelocity
0.5 0 0
1
2
Time [s]
Time [s]
Fig. 8. App1: Power Control with pure mo- Fig. 9. App1: Power control with impedance Fig. 10. App1: Power control with vari-
0 −0.5 1
1.5
2
2.5
3
2
L2R R2L
1 0 0.5
1
1.5
2
2.5
E2port ∗ ESt
0 0.5
1
1.5
2
2.5
0.5 0 0.5
µForce νVelocity 1
1.5
2
2.5
3
3
3.5
4
4.5
5
5.5
Position [rad] 6
0
−2 6 1
2
2.5
3
3.5
4
4.5
5
5.5
6
L2R R2L
2 1 0 2
2.5
3
3.5
4
4.5
5
5.5
6
0.2
E2port ∗ ESt
0.1 0 2
2.5
3
3.5
4
4.5
5 µForce 5.5
6
νVelocity
1 0.5
Master Slave 8
10
12
14
16
8
10
12
14
16
10
12
14
16
10
12
14
16
10
12
14
16
12
14
16
0.5 0
−0.5 6 10
−1
3
1
2.5
−0.5
3
0.2 0.1
2 0.5
Scaling [−]
Scaling [−]
Energy [J ]
Power [J /s]
−1 0.5
P osition [rad]
3
L2R Energies Torque [Nm] [J ]
2.5
Scaling [−] Scaling [−]
2
0 −1
5
E2L2R E3L2R E4L2R
0 6 1
8
0.5 0 6 1
ν PC ν∗ 8
0.5
0 6 0.05
νVelocity 8
4 −5
1.5
Master Slave
E2port [J ]
1
Torque [Nm]
0.5
3 2 1 0 −1
Power [J /s]
0.5
Master Slave
able motion scaling and velocity scaling adaptation.
scaling and force scaling adaptation. textfill next line
Energy [J ]
P osition [rad]
2 1 0 −1
Torque [Nm]
tion scaling and velocity scaling adaptation. textfill next line
0 2
2.5
3
3.5
4
4.5
5
5.5
6
0 6
8
10
Time [s]
Time [s]
Time [s]
Fig. 11. App2: Energy control with motion Fig. 12. App2: Energy control with force Fig. 13. TDPA with Approach 1 and scaling and velocity scaling adaptation. textfill next line
velocity scaling adaptation at 100ms roundtrip-delay
scaling and force scaling adaptation. textfill next line
In the second experiment (see Fig. 9), an impedance scaling (ν des = 0.5, µdes = 2) with force scaling adaptation (5) is presented. The master moves the slave t = [1.5s − 4.3s] and the slave overtakes the lead at t = [4.3s − 6s]. Comparing the power and scaling plots, it is obvious that the force scaling is adapted when energy flows from slave to master (R2L direction, P1L2R = 0). It can be seen from the energy plot that also the force adaptation leads to a passive scaling 2-port.
∗
ν =
ν des
ν
, if |vm | < vmin + , if vmin ≤ |vm | < 2 rad m |−vmin s (1 − ν des ) v|vmax −vmin rad 1.2 , if |vm | ≥ 2 s .
des
The variable scaling was designed such that at fast motions, the velocity scaling is ν = 1.2. If ν > 1 (t = [4.8s − 5.2s]), the time domain scaling controller has to set the force scaling to µ = ν independent of the power flow to maintain passivity. High master velocities appear at free motions and therefore, the increase of force scaling has no unintuitive effect. If the controller has to act, the velocity scaling is set to ν = µ (t = [5.5s − 6.8s]). Again,
The third experiment in Fig. 10 presents a variable motion scaling and power-based velocity scaling adaptation (4). The velocity scaling depends on the master velocity ν = f (vm ):
59
2019 IFAC TDS 36 Sinaia, Romania, September 9-11, 2019
M. Panzirsch et al. / IFAC PapersOnLine 52-18 (2019) 31–36
Proceedings. 1998 IEEE International Conference on Robotics and Automation, volume 2, 1609–1614. IEEE. Itoh, T., Kosuge, K., and Fukuda, T. (2000). Humanmachine cooperative telemanipulation with motion and force scaling using task-oriented virtual tool dynamics. IEEE Transactions on robotics and automation, 16(5), 505–516. Jazayeri, A. and Tavakoli, M. (2013). Absolute stability analysis of sampled-data scaled bilateral teleoperation systems. Control Engineering Practice, 21(8), 1053– 1064. Khan, S., Sabanovic, A., and Nergiz, A.O. (2009). Scaled bilateral teleoperation using discrete-time sliding-mode controller. IEEE Transactions on Industrial Electronics, 56(9), 3609–3618. Niemeyer, G.D. (1996). Using wave variables in time delayed force reflecting teleoperation. Ph.D. thesis, Massachusetts Institute of Technology. Onal, C.D. and Sitti, M. (2009). A scaled bilateral control system for experimental one-dimensional teleoperated nanomanipulation. The International Journal of Robotics Research, 28(4), 484–497. Panzirsch, M., Artigas, J., Ryu, J.H., and Ferre, M. (2013). Multilateral control for delayed teleoperation. In Proc. of 2013 International Conference on Advanced Robotics, 1–6. Panzirsch, M., Balachandran, R.R., Weber, B., Ferre, M., and Artigas, J. (2018). Haptic augmentation for teleoperation through virtual grasping points. IEEE Transactions on Haptics, 11(3), 400–416. Panzirsch, M., Ryu, J.H., and Ferre, M. (2019). Reducing the conservatism of the time domain passivity approach through consideration of energy reflection in delayed coupled network systems. Mechatronics, 58, 58–69. Ryu, J.H., Artigas, J., and Preusche, C. (2010). A passive bilateral control scheme for a teleoperator with timevarying communication delay. Mechatronics, 20(7), 812– 823. Secchi, C., Straimgioli, S., and Fantuzzi, C. (2005). Power scaling in port-hamiltonian based telemanipulation. In 2005 IEEE/RSJ International Conference on Intelligent Robots and Systems, 1850–1855. IEEE. Speich, J.E. and Goldfarb, M. (2002). Implementation of loop-shaping compensators to increase the transparency bandwidth of a scaled telemanipulation system. In Proceedings 2002 IEEE International Conference on Robotics and Automation, volume 3, 2886–2893. IEEE. Vander Poorten, E.B., Yokokohji, Y., and Yoshikawa, T. (2006). Stability analysis and robust control for fixedscale teleoperation. Advanced Robotics, 20(6), 681–706. Yan, J. and Salcudean, S.E. (1996). Teleoperation controller design using h-infinity optimization with application to motion-scaling. IEEE Transactions on control systems technology, 4(3), 244–258.
the energy plot E2port confirms the passivity of the time domain scaling controller. Figure 11 and Fig. 12 present the energy-based Approach 2 with velocity (6) and force scaling adaptation (7) respectively. The passivity of the scaling is confirmed by the solid ∗ energy plot E2port . The dashed energy plot ESt presents the reference energy which is reset when the positions of master and slave match. The master leads in Fig. 11 at t = [0.5s − 2.25s] and in Fig. 12 at t = [2.1s − 4.6s]. The positive aspect of Approach 2 is that the scaling does not change during the wall contact in contrast to the powerbased approaches. The experiment in Fig. 13 presents a delayed setup with TDPA at 100ms roundtrip-delay with power-based scaling (Approach 1) and velocity scaling adaptation (4). A pure motion scaling (ν des = 0.7, µdes = 1) has been applied. Since E4L2R ≤ E2L2R (compare Fig. 7) the energy plot confirms that the admittance type PC ensures a passive communication. The scaling ν ∗ acts on the velocity output of the PC. The resulting velocity scaling ν is ν P C when ν P C < ν des and ν des when ν P C > ν des . When power flows from slave to master in R2L direction, the velocity scaling is deactivated (ν = 1). The consideration of the PC dissipation in the motion scaling leads to lower conservatism. 5. CONCLUSION A new time domain passivity control for scaling subsystems has been proposed that focuses on setups that require a down-scaling of motions for higher precision or a smaller workspace of the slave robot. A power- and an energy-based control approach have been presented that can be applied for impedance and pure motion scaling with adaptation of force or velocity. The user has to decide for a force or velocity adaptation according to the respective teleoperation application. A time delay control approach has been extended with the proposed scaling method. The time domain control approach has been successfully validated in experiments. In contrast to former approaches, in the proposed method, a variable scaling is feasible, which is otherwise very difficult to integrate in an alternative frequency based stability analysis. Furthermore, no passivity assumptions for environments have to be made. REFERENCES Boukhnifer, M. and Ferreira, A. (2006). Stability and transparency for scaled teleoperation system. In 2006 IEEE/RSJ International Conference on Intelligent Robots and Systems, 4217–4222. IEEE. Boukhnifer, M., Ferreira, A., and Fontaine, J.G. (2004). Scaled teleoperation controller design for micromanipulation over internet. In IEEE International Conference on Robotics and Automation, 2004. Proceedings. ICRA’04. 2004, volume 5, 4577–4583. IEEE. Colgate, J.E. (1991). Power and impedance scaling in bilateral manipulation. In Proceedings. 1991 IEEE International Conference on Robotics and Automation, 2292–2297. IEEE. Goldfarb, M. (1998). Dimensional analysis and selective distortion in scaled bilateral telemanipulation. In 60