Adaptive Implicit Hybrid Force/Pose Control of ... - Semantic Scholar
Proceedings of 2004 IEEE/RSJ International Conference on Intelligent Robots and Systems September 28 - October 2, 2004, Sendai, Japan
Adaptive Implicit Hybrid Force/Pose Control of Industrial Manipulators: Compliant Motion Experiments Torsten Kr¨oger, Bernd Finkemeyer, Markus Heuck, and Friedrich M. Wahl Institute for Robotics and Process Control Technical University of Braunschweig, Germany {t.kroeger, b.finkemeyer, m.heuck, f.wahl}@tu-bs.de
Abstract— The major purpose of this paper is to combine results of current robot force control research with scientific approaches in compliant motion, which are based on Mason’s Task Frame Formalism. The embedding of adaptive implicit hybrid force/pose control in a robot control architecture for compliant motion control is described. By the usage of adaptive force control, the practicability of compliant motion applications is improved. The applied control concept is constituted in a theoretical as well as in a practical manner. To highlight the meaning for practical implementations, experimental results with industrial manipulators under adaptive force control in three degrees of freedom are finally shown.
I. I NTRODUCTION Compliant motion of robot manipulators occurs when the manipulator motion is constrained by environmental objects. M. T. Mason was one of the first, who published works on compliance and force control [1]. The abstract high level commands specified by Mason’s Task Frame Formalism (TFF) enable robot application programmers to construct solutions in a very intuitive way. It is the task of the robot control system to transform these high level commands to unambiguous low level control statements. Robot control architectures offering this type of functionality need to provide at least hybrid force/pose control, which is an essential part of the TFF. Potential application fields are assembly, deburring, welding, disassembly, machining, essential contact tasks in general. Since Mason, many research groups published approaches in this field. Raibert, and Craig [2] presented one of the first hybrid position/force control concepts. A huge variety of approaches can be found in the open literature, e.g. [3]– [5]. Up to now research groups are dealing with force control research, but the basic concepts do not change anymore. The latest works of Whitcomb [6], [7] contain approaches of adaptive force control for industrial manipulators in theory and experiment. De Schutter, Van Brussels, and Bruyninckx delivered significant articles about research in compliant motion specification [8]–[10] and compliant motion control [11]. Schimmels and Huang wrote about force-guided assembly in a very theoretical manner [12]. The PhD thesis of Natale [13] constitutes the objective in clear theoretical expressions as well as in good practical experiments. Our institute works on
0-7803-8463-6/04/$20.00 ©2004 IEEE
research in assembly planning [14] as well as on robot control [15], [16]. The long-term aim is to bring both fields together [17]. One major part in this challenge is to develop a robot control architecture, which provides at least hybrid force/pose control, to execute TFF manipulation primitives. This paper focusses on the force control part within a compliant motion control scheme. Because of the generality and the wide field of possible application, a respective force controller must be able to act in a huge variety of environments to handle different tasks. I.e. the dynamic parameters of the environment might change permanently and are not prior known by the (end) user. A force controller, whose parameters are set up for a fixed environment causes non-optimal behavior, because it is designed for one specific environment. Embedding adaptive force/pose control for TFF experiments is supposed to deliver better results in practice, since the system is able to react on changing dynamic behaviors. An adaptive force control scheme for the three translational degrees of freedom (DOF) is introduced and experimental results are given in this paper. By the best of our knowledge, this is the first approach, which delivers experimental results for adaptive force control in three DOFs for executing compliant motion tasks. Section II introduces conventions and the applied nomenclature used in section III to describe the control scheme, the identification algorithm as well as the adaption method. To highlight the meaning for practical implementations, this paper emphasizes on experimental results; simulation results are only cited. The essential parts of our experimental setup are given in section IV. Section V discusses the substantial results, while section VI concludes with a short summary and suggestions for future work. II. N OMENCLATURE AND C ONVENTIONS This chapter is supposed to introduce the applied nomenclature and important conventions used in this paper. In general, the TFF allows multi-sensor integration, but for a better understanding, we only consider force/torque sensors. As can be seen in Fig. 1, the frame hand Tsensor represents the sensor location with respect to the robot hand frame, whose pose depends on the robot kinematic given by robot Thand .
816
task
world
Ttask =
world T−1 Ttask =⇒ task ·
task
P~ task (3)
Angle brackets denote desired values, e.g. hworld P~ task i contains the desired pose w.r.t. the world frame (WF). The desired poses are specified by the user by a sequence of single manipulation primitives. III. C ONTROL A RCHITECTURE
Fig. 1.
Frame assignments for the proposed control scheme
robot
robot ~ hand
P
=
Thand ⇔ robot P~
hand
(1)
(robot pxhand ,robot pyhand ,robot pzhand , robot
ϕxhand ,robot
ϕyhand ,robot
I=
ϕzhand )
(2) Frames can be mapped to position vectors, such that equation 1 can be considered as bijective. T ǫ 4×4 is a homogeneous coordinate transformation, P~ ǫ 6 represents the corresponding pose vector. When talking about “pose”, all six DOFs are meant, i.e. the “position”, p~ ǫ 3 , and the “orientation”, ϕ ~ ǫ 3 . Analogous to this, a force/torque vector F~ ǫ 6 is composed by a force f~ ǫ 3 and a torque ~τ ǫ 3 . When talking about “force control”, only the translational DOFs are meant. “Torque control” addresses the rotational DOFs. The orientation of any pose vector is supposed to be given in roll-pitch-yaw (RPY) angles. All desired values given by the user are assumed to be given w.r.t. the task frame (TF). The TF is user-given and is described by a fixed transformation w.r.t. the robot hand frame. I.e. to perform force/torque control w.r.t. the TF, the measured force/torque values from the sensor frame have to be transformed into the hand frame and subsequently into the TF. Regarding differential orientation changes, which appear if the pose difference from one control cycle to the next one is considered, the following nomenclature is introduced: When executing a movement with respect to the TF and calculating new values in a control cycle, we refer to the old ones from the last control cycle. To distinguish both, values from the last control cycle are underlined. E.g., the task frame displacement w.r.t. the TF of the last cycle is represented by task Ttask or in vector notation task P~ task , where the translational x. component is denoted by task ptask x
R
R
R
Robot force/torque control can be categorized in impedance control and hybrid force/torque/pose control. In impedance control, a prescribed static or dynamic relation between the robot and the effector force is supposed to be maintained, e. g. [3]. Hybrid control can be divided into two further subcategories: explicit and implicit hybrid control. Explicit hybrid force/torque control architectures directly command the joint torques by the sensed force/torque error, e. g. [1], [2]. In implicit hybrid force/torque control, the inner joint position/velocity control loop is cascaded by an outer force control loop, as proposed in [11]. The here-proposed adaptive implicit hybrid control scheme is shown in Fig. 2 and is based on an architecture described in [16]. The user-given selection matrices S0 . . . Sn determine, which DOF is controlled by which controller, where n − 1 is the number of installed controllers.
R R
R
R
n−1 X
Sl
(4)
l=0
S represents the identity matrix. In the special case of this paper, we only consider pose (S0 ) and force/torque (S2 ) control. S1 is reserved for velocity control. The elements of the diagonal matrices are sii ǫ [0, 1]. To inhibit force control in free space and to prevent from sensor overload, we introduce the matrix Ξ2 , which is also a diagonal matrix: Its elements are defined as
ξii =
1 :
task Fi > task
Fi
Fi i = sign
∧ sign h ∨ htask Fi i < ∧ (|sensor Fi |
Adaptive Implicit Hybrid Force/Pose Control of Industrial Manipulators: Compliant Motion Experiments Torsten Kr¨oger, Bernd Finkemeyer, Markus Heuck, and Friedrich M. Wahl Institute for Robotics and Process Control Technical University of Braunschweig, Germany {t.kroeger, b.finkemeyer, m.heuck, f.wahl}@tu-bs.de
Abstract— The major purpose of this paper is to combine results of current robot force control research with scientific approaches in compliant motion, which are based on Mason’s Task Frame Formalism. The embedding of adaptive implicit hybrid force/pose control in a robot control architecture for compliant motion control is described. By the usage of adaptive force control, the practicability of compliant motion applications is improved. The applied control concept is constituted in a theoretical as well as in a practical manner. To highlight the meaning for practical implementations, experimental results with industrial manipulators under adaptive force control in three degrees of freedom are finally shown.
I. I NTRODUCTION Compliant motion of robot manipulators occurs when the manipulator motion is constrained by environmental objects. M. T. Mason was one of the first, who published works on compliance and force control [1]. The abstract high level commands specified by Mason’s Task Frame Formalism (TFF) enable robot application programmers to construct solutions in a very intuitive way. It is the task of the robot control system to transform these high level commands to unambiguous low level control statements. Robot control architectures offering this type of functionality need to provide at least hybrid force/pose control, which is an essential part of the TFF. Potential application fields are assembly, deburring, welding, disassembly, machining, essential contact tasks in general. Since Mason, many research groups published approaches in this field. Raibert, and Craig [2] presented one of the first hybrid position/force control concepts. A huge variety of approaches can be found in the open literature, e.g. [3]– [5]. Up to now research groups are dealing with force control research, but the basic concepts do not change anymore. The latest works of Whitcomb [6], [7] contain approaches of adaptive force control for industrial manipulators in theory and experiment. De Schutter, Van Brussels, and Bruyninckx delivered significant articles about research in compliant motion specification [8]–[10] and compliant motion control [11]. Schimmels and Huang wrote about force-guided assembly in a very theoretical manner [12]. The PhD thesis of Natale [13] constitutes the objective in clear theoretical expressions as well as in good practical experiments. Our institute works on
0-7803-8463-6/04/$20.00 ©2004 IEEE
research in assembly planning [14] as well as on robot control [15], [16]. The long-term aim is to bring both fields together [17]. One major part in this challenge is to develop a robot control architecture, which provides at least hybrid force/pose control, to execute TFF manipulation primitives. This paper focusses on the force control part within a compliant motion control scheme. Because of the generality and the wide field of possible application, a respective force controller must be able to act in a huge variety of environments to handle different tasks. I.e. the dynamic parameters of the environment might change permanently and are not prior known by the (end) user. A force controller, whose parameters are set up for a fixed environment causes non-optimal behavior, because it is designed for one specific environment. Embedding adaptive force/pose control for TFF experiments is supposed to deliver better results in practice, since the system is able to react on changing dynamic behaviors. An adaptive force control scheme for the three translational degrees of freedom (DOF) is introduced and experimental results are given in this paper. By the best of our knowledge, this is the first approach, which delivers experimental results for adaptive force control in three DOFs for executing compliant motion tasks. Section II introduces conventions and the applied nomenclature used in section III to describe the control scheme, the identification algorithm as well as the adaption method. To highlight the meaning for practical implementations, this paper emphasizes on experimental results; simulation results are only cited. The essential parts of our experimental setup are given in section IV. Section V discusses the substantial results, while section VI concludes with a short summary and suggestions for future work. II. N OMENCLATURE AND C ONVENTIONS This chapter is supposed to introduce the applied nomenclature and important conventions used in this paper. In general, the TFF allows multi-sensor integration, but for a better understanding, we only consider force/torque sensors. As can be seen in Fig. 1, the frame hand Tsensor represents the sensor location with respect to the robot hand frame, whose pose depends on the robot kinematic given by robot Thand .
816
task
world
Ttask =
world T−1 Ttask =⇒ task ·
task
P~ task (3)
Angle brackets denote desired values, e.g. hworld P~ task i contains the desired pose w.r.t. the world frame (WF). The desired poses are specified by the user by a sequence of single manipulation primitives. III. C ONTROL A RCHITECTURE
Fig. 1.
Frame assignments for the proposed control scheme
robot
robot ~ hand
P
=
Thand ⇔ robot P~
hand
(1)
(robot pxhand ,robot pyhand ,robot pzhand , robot
ϕxhand ,robot
ϕyhand ,robot
I=
ϕzhand )
(2) Frames can be mapped to position vectors, such that equation 1 can be considered as bijective. T ǫ 4×4 is a homogeneous coordinate transformation, P~ ǫ 6 represents the corresponding pose vector. When talking about “pose”, all six DOFs are meant, i.e. the “position”, p~ ǫ 3 , and the “orientation”, ϕ ~ ǫ 3 . Analogous to this, a force/torque vector F~ ǫ 6 is composed by a force f~ ǫ 3 and a torque ~τ ǫ 3 . When talking about “force control”, only the translational DOFs are meant. “Torque control” addresses the rotational DOFs. The orientation of any pose vector is supposed to be given in roll-pitch-yaw (RPY) angles. All desired values given by the user are assumed to be given w.r.t. the task frame (TF). The TF is user-given and is described by a fixed transformation w.r.t. the robot hand frame. I.e. to perform force/torque control w.r.t. the TF, the measured force/torque values from the sensor frame have to be transformed into the hand frame and subsequently into the TF. Regarding differential orientation changes, which appear if the pose difference from one control cycle to the next one is considered, the following nomenclature is introduced: When executing a movement with respect to the TF and calculating new values in a control cycle, we refer to the old ones from the last control cycle. To distinguish both, values from the last control cycle are underlined. E.g., the task frame displacement w.r.t. the TF of the last cycle is represented by task Ttask or in vector notation task P~ task , where the translational x. component is denoted by task ptask x
R
R
R
Robot force/torque control can be categorized in impedance control and hybrid force/torque/pose control. In impedance control, a prescribed static or dynamic relation between the robot and the effector force is supposed to be maintained, e. g. [3]. Hybrid control can be divided into two further subcategories: explicit and implicit hybrid control. Explicit hybrid force/torque control architectures directly command the joint torques by the sensed force/torque error, e. g. [1], [2]. In implicit hybrid force/torque control, the inner joint position/velocity control loop is cascaded by an outer force control loop, as proposed in [11]. The here-proposed adaptive implicit hybrid control scheme is shown in Fig. 2 and is based on an architecture described in [16]. The user-given selection matrices S0 . . . Sn determine, which DOF is controlled by which controller, where n − 1 is the number of installed controllers.
R R
R
R
n−1 X
Sl
(4)
l=0
S represents the identity matrix. In the special case of this paper, we only consider pose (S0 ) and force/torque (S2 ) control. S1 is reserved for velocity control. The elements of the diagonal matrices are sii ǫ [0, 1]. To inhibit force control in free space and to prevent from sensor overload, we introduce the matrix Ξ2 , which is also a diagonal matrix: Its elements are defined as
ξii =
1 :
task Fi > task
Fi
Fi i = sign
∧ sign h ∨ htask Fi i < ∧ (|sensor Fi |
