SENSORS, IDENTIFICATION, AND LOW LEVEL

September 28, 2004 19:59 WSPC/191-IJHR

00025

Int. J. Human. Robot. 2004.01:517-532. Downloaded from www.worldscientific.com by KARLSRUHE INSTITUTE OF TECHNOLOGY (KIT) KIT LIBRARY on 11/13/12. For personal use only.

International Journal of Humanoid Robotics Vol. 1, No. 3 (2004) 517–532 c World Scientific Publishing Company

SENSORS, IDENTIFICATION, AND LOW LEVEL CONTROL OF A FLEXIBLE ANTHROPOMORPHIC ROBOT HAND

JAN MARTIN∗ , SEBASTIAN BECK, ARNE LEHMANN, RALF MIKUT, CHRISTIAN PYLATIUK, STEFAN SCHULZ and GEORG BRETTHAUER Forschungszentrum Karlsruhe GmbH Institute for Applied Computer Science P.O. Box 3640, 76021 Karlsruhe, Germany ∗[email protected] www.iai.fzk.de Received 17 April 2004 Revised 15 June 2004 Accepted 15 June 2004 The successful control of a robot hand with multiple degrees of freedom not only requires sensors to determine the state of the hand but also a thorough understanding of the actuator system and its properties. This article presents a set of sensors and analyzes the actuator properties of an anthropomorphic robot hand driven by flexible fluidic actuators. These flexible and compact actuators are integrated directly into the finger joints, they can be driven either pneumatically or hydraulically. The sensors for the measurement of joint angles, contact forces, and fluid pressure are described; the designs utilize mostly commodity components. Hall sensors and customized half-ring rare-earth magnets are used to integrate the joint angle sensors directly into the actuated joints. A force sensor setup allowing soft finger surfaces is evaluated. Fluid pressure sensors are needed for the model-based computation of joint torques and to limit the actuator pressure. Static and dynamic actuator characteristics are determined in a theoretical process analysis, and suitable parameters are identified in several experiments. The resulting actuator model incorporates the viscoelastic material behavior and describes the relations of joint angle, actuator pressure, and actuator torque. It is used in simulations and for the design of a joint position controller. Keywords: Anthropomorphic robot hands; flexible fluidic actuators; sensors; control.

1. Introduction Artificial hands are an essential contribution to enable a humanoid robot with real-world interaction and cooperation. Direct interaction with humans and human environments poses several challenges to the design of suitable artificial hands. We have earlier presented an approach for such an anthropomorphic, light-weight five-fingered hand with inherent compliance using flexible fluidic actuators integrated into the finger joints.1–4 These actuators are closed elastic chambers; they can be inflated either pneumatically or hydraulically resulting in a joint torque. 517



518

00025

J. Martin et al.

Fig. 1. Concept for an integrated hydraulic five-finger-hand.

Our hand prototype incorporates an aluminum skeleton with 11 DOF, each finger has two joints for flexion; the thumb is equipped with a third joint to allow opposition with other fingers. Previous sensorless prototypes of this hand (Fig. 1) already allowed the successful, time-controlled execution of various prehensile patterns like cylindrical grasp, tripod grasp and lateral grasp. Improvements in grasping and manipulation using this actuator technology require sensors and advanced control strategies. In the following section, we will present a set of sensors compliant with our hand design to obtain feedback on the hand’s state and its interaction with the environment. The last part of this paper covers an actuator model describing the actuator properties relevant for position and torque control as well as the controller itself. 2. Sensor System Three classes of sensors are considered in this paper: joint angle, contact force and fluid pressure. Besides compatibility with the existing hand concept, the development focused mainly on simplicity and the use of reliable commodity components. 2.1. Joint angle sensors Many researchers have already addressed the issue of finger joint angle measurement. Depending on the chosen actuator principles, either finger-integrated or external sensors were chosen. Typical integrated position sensors were conductive plastic potentiometers,5 optical position sensing devices6 or hall-sensors with permanent magnets mounted to a slider mechanism.7,8 In all these cases, the actuators were not directly integrated into the finger joints but into the phalanges. Especially for fingers driven externally via tendons9 or ball-spindle drives,10 DC motors with optical


00025


Sensors, Identification, and Control of a Robot Hand

Air Pressure Supply

Pressure Control Valve

Control Unit

Valve Control

519

Flexible Actuator

Phalanx Angle Magnet Joint Axis

ϕ

Hall Sensor Phal

anx

Fig. 2. System structure of a controlled finger joint. Left: pressure supply, valve and control unit. Right: arrangement of phalanges, actuator, magnet and hall sensor within a finger joint (not to scale).

encoders were chosen. Such external sensors typically did not have to satisfy hard package constraints. For finger joints with integrated flexible fluidic actuators, a sensor fitting directly into the joint mechanism with a flexion axis close to the palmar side of the fingers (Fig. 2) was needed. The developed sensor relies on two main principles: • the magnitude of flux density along the perimeter of a circular permanent magnet with diametrical magnetization is constant; • the output signal of a Hall sensor is sensitive to the direction of the magnetic field and proportional to the normal flux density. A Hall sensor to measure the normal flux density Bn of a rotating diametrically magnetized disc magnet at a constant position results in a sinusoidal signal as in Eq. (1). The signal has a maximum normal flux density B0 along the magnetization axis and an offset angle ϕ0 between the magnetization axis and the hall element’s normal axis: Bn = B0 · sin(ϕ − ϕ0 ).

(1)

With the finger joint having a range of motion of −10◦ ≤ ϕ ≤ +100◦ and ϕ0 = 45◦ , we obtain a signal as in Fig. 3. To minimize the effects on joint structure stability, the magnet is reduced to a half-ring shape resulting in a setup and signal as in Fig. 4. As a consequence, a lower signal amplitude and an increased nonlinearity at extreme angles is obtained. The sensor signals from simulations and prototypes were approximated with a first-order polynomial equation. The sensor nonlinearity l was then evaluated using Eq. (2) as suggested in Ref. 7 with n measured signal values Um and approximated values Ul giving the results in Table 1: n 2 100% i=1 (Um,i − Ul,i ) · . (2) l= |Um,n − Um,1 | n


520

00025

J. Martin et al.

0.1

B [T]

0.05

0.05

0.1 90

45

0

45

90 ϕ [˚]

135

180

225

270

Fig. 3. 2-D FEM magnet simulation setup (left) and output signal (right) of a hall sensor moved along the perimeter of a ring magnet with ϕ0 = 45◦ .

half ring magnet ring magnet

0.1

0.05 B [T]

0

n


n

0

0.05

0.1 90

45

0

45

90 ϕ [˚]

135

180

225

270

Fig. 4. 2-D FEM magnet simulation setup (left) and output signal (right) of a hall sensor moved along the perimeter of a half-ring magnet with ϕ0 = 45◦ .

Table 1. Measured sensor signal linearity l. Simulation Joint range of motion Ring magnet Half-ring magnet Slider mechanism7

0◦

· · · 90◦

0.014% 0.039%

−10◦

Measurement · · · 100◦

0.039% 0.117%

0◦

· · · 90◦

0.058% 0.098%

−10◦ · · · 100◦ 0.280% 0.317%

3.8–5.3%

As expected, the half-ring magnet sensor version performs worse than the full-ring version and the actual prototypes perform worse than the corresponding simulation. However, even the half-ring version presents a more than 10 times lower nonlinearity than the comparable slider mechanism while being used over an extended range of motion (−10◦ · · · 100◦ ).


00025

521

The employed ratiometric hall sensor with an integrated amplifier and a temperature compensation provides an output signal Uout as in Eq. (3), which can be fed directly into an analog/digital converter. U0 denotes the sensor supply voltage, and Bmax is the maximum flux density that can be handled by the amplifier: Bn U0 · 1+ Uout = . (3) 2 2 · Bmax Output signals covering the range from approximately 0.5 V to 4.5 V can be obtained using rare-earth NdFeB magnets and a suitably chosen distance between magnet and hall sensor. An optimization of the signal linearity was performed using a genetic algorithm implemented in Matlab and FEMM11 as a 2D-FEM magnet simulation tool. The steady-state genetic algorithm derived from Ref. 12 with a population size of 25 uses tournament selection. Its genotypes describe the radius of the magnet contour in steps of 10◦ . A smooth magnet contour is obtained by computing the in-between points in steps of 1◦ using: n (βi −ϕ)2 r(ϕ) = r0 + gi · e (2·σ)2 . (4) i=1

As expected, we see an increased magnet radius in the outer areas reducing the distance between the magnet and the hall sensor resulting in a higher measured flux density (Fig. 5, left). The resulting magnet contour is shown in Fig. 5, right. magnet radius

magnet contour

2.8 2 2.6

y [mm]

r [mm]



2.4

1 0 –1 –2

2.2 – 45

0

45 α [˚]

90

135

–2 –1 0 1 x [mm]

2

Fig. 5. Geometry optimization results: magnet radius (left) and magnet contour (right).

Figure. 6 demonstrates the optimized signal curve and the position error estimated from the simulation results. Such a linear sensor signal simplifies the calibration process for newly assembled joints to a linear interpolation between the values measured at the two end positions (0◦ and 90◦ ). 2.2. Contact force sensors Force sensors13,14 are essential for the detection of object contact during grasping. Numerous types have been used in artificial hands ranging from binary switches


522

00025

J. Martin et al.

normal flux density

position error 0.5

90 δ [˚]

estimated ϕ [˚]


180

0

0 –0.5 –90

– 0.1 – 0.05 0 0.05 0.1 Bn [T]

0

45 estimated ϕ [˚]

90

Fig. 6. Sensor signals after optimization: normal flux density (left) and position error (right).

to strain gages, conductive polymer film sensors, hall sensors15 and tactile arrays. For the use in our hand, force sensors had to fulfill three central criteria: minimized thickness to fit between the bone structure and the palmar hand surface, compliance with a soft palm to increase grasp safety16 and a high resolution for forces below 5 N. Among the conductive polymer sensors, force sensing resistors (FSR, Interlink Electronics) are a well-known and cheap type of sensor not only used in robotics but also in force measurement on humans in programming by demonstration environments17 and physiology.18,19 These sensors are made of three coated plastic film layers giving an overall thickness of about 0.5 mm. The bottom layer is coated with a semiconductive layer; the top layer is printed with two interdigitated conducting structures. A spacer layer in the outer sensor areas separates these two layers. We found a FSR with an outer diameter of 7.62 mm (type FSR-149, Fig. 7, right) to be the most suitable for the integration into the fingers of a hand approximating human size. The basic sensor characteristic approximates a negative linear relationship between the logarithms of the applied force F and the resistance RS (Eq. (5) and Fig. 7, left): log RS ≈ c0 − c1 · log F .

(5)

This nonlinear relationship can be exploited with a voltage divider setup (Fig. 8, left) measuring the output voltage UD at the fixed resistor RD . A high sensitivity is obtained in the force range between 0 and 5 N while larger forces only result in small signal changes (see Fig. 8). The best way to compute the applied force from the voltage divider output signal was to determine the FSR resistance from the supplied voltage and the fixed resistor RD and to approximate Eq. (5) with a third-order polynomial using leastsquares regression. The calibration data was obtained from a test rig with a digital lab scale as a reference force sensor. The issues of well-defined actuation, sensing of small contact forces, mechanical sensor protection against sharp edges, and a soft finger surface are resolved


00025


523

6

resistance RS [Ω]

10

measurement approximation

5

4

10 –1 10

0

1

10

2

10

10

force [N] Fig. 7. Left: force-resistance diagram of FSR sensor. Right: FSR sensor, inside view of silicone covers for phalanges and finger, phalanx with FSR sensor and silicone cover.

3

RS U0 RD

+

UD

voltage U D [V]


10

2

1 measurement approximation 0

0

5

10 force [N]

15

Fig. 8. FSR force sensor in a voltage divider setup. Left: schematic with unity gain buffer circuit. Right: force-voltage diagram.

with sensor covers as depicted in Fig. 7, right. A prototype aluminum mold is used to manufacture the sensor covers from medical two-component silicone (Detax Durosil). The sensor is actuated with a sphere of 4 mm diameter. For the fingertips, the outer cylindrical shape ends with a spherical fingertip. For physiological measurements, this actuation principle was modified for the attachment to human and prosthetic fingers.19 2.3. Actuator pressure sensor setup In a fluidically driven artificial hand, pressure sensors are essential for a number of functions such as position control, torque estimation, and monitoring to prevent actuator damage. Tests have shown that for pneumatics a pressure measurement in or near the actuators is not necessary for our requirements. This allows one to integrate the pressure sensors into future valve units in the hand’s palm with less strict package requirements and easier wiring.


524

00025

J. Martin et al.

3. Modeling and Low-Level Control In this section, an actuator model is derived from theoretical analysis and measurements performed on prototype parts. A pneumatic joint position controller is deduced from this actuator model, followed by considerations on future controller developments. 3.1. Static actuator characteristics To obtain the static actuator torque characteristics Ma (ϕ, pa ), static experiments were performed in a test rig. The joint torque is measured using an industrialgrade torque sensor for the given joint angles ϕ set by a step motor and actuator pressures pa set with a pneumatic pressure-control valve. The values obtained are shown in Fig. 9, left. Multilinear regression was used to compute the parameters a0 to a3 in Eq. (6) used later as a static actuator model (Fig. 9, left): Ma (pa , ϕ) = a0 + a1 · ϕ + a2 · pa + a3 · pa · ϕ.

(6)

During these experiments, we also determined the actuator volume characteristics Va (ϕ, pa ). 0.6

0.6

0.5

0.5

0.4

0.4

p =5*105 Pa a

M (ϕ,p ) [Nm]

p =5*105 Pa a

0.3

a

0.3 0.2

a

M a (ϕ,p a ) [Nm]


We will see in Sec. 3 that in pneumatic applications, the joint position should be controlled by setting an actuator pressure, so this method requires a low-level pressure control loop which is accomplished with an analog pressure control valve. For our type of actuators, the actuator torque can be estimated from the measured actuator pressure and the joint angle.

0.1

0.1

0 –0.1

0.2

0 p a =0 0

10

20

30

40 50 ϕ[˚]

60

70

80

90

–0.1

p a =0 0

10

20

30

40 50 ϕ[˚]

60

70

80

Fig. 9. Measured (left) and approximated (right) static actuator torque characteristics.

90


00025


525

The previously described torque behavior is independent of the chosen fluid. However, the stiffness properties vary significantly between pneumatics and hydraulics. Torque trajectories of an actuator filled with constant fluid amounts were computed from the torque Ma (ϕ, pa ) and volume Va (ϕ, pa ) characteristics. In the case of pneumatics (Fig. 10, left), we approximate the fluid behavior using Boyle–Mariotte’s law for a constant mass m (i.e. a closed valve) of an ideal gas at constant temperature T : pa · Va = const. |T =const.,m=const.

(7)

Here, a soft behavior can be observed when opening the finger with an external torque. A large portion of the trajectories reaches ϕ = 0◦ with a moderately increased pressure. For hydraulics (Fig. 10, right) and the chosen pressure range, we can assume an incompressible fluid with a pressure-independent volume, i.e. any observed compliance depends solely on the actuator. Here, the system behaves much more stiffly and we can see that most trajectories reach the maximum actuator pressure (6 × 105 Pa) before the joint is fully opened. Such a compliant actuator behavior can be considered favorable during grasping, as the fingers will follow object position disturbances without controller activity or an immediate loss of object contact.

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 – 0.1 – 0.2 0

M a [Nm]

0.7 0.6 0.5 0.4 0.3 0.2 0.1 0 – 0.1 – 0.2 0

a

M [Nm]


3.2. Actuator compliance

10

20

30

40

50

60

ϕ [˚]

70

80

90

0

1

2

4

3 5

p a [10 Pa]

5

6

10

20

30

40

50

60

ϕ [˚]

70

80

90

0

1

2

4

3

5

6

5

p a [10 Pa]

Fig. 10. Torque trajectories of constant fluid masses for pneumatics (left) and hydraulics (right) computed from experimental results.

3.3. Dynamic behavior More effects have to be taken into consideration for the dynamic properties of the actuator. From assuming a model structure (Fig. 11) with a speed-proportional friction d, an inertia J of the distal phalanx and a disturbance torque Md , we obtain the differential equation: Ma (pa , ϕ) = J · ϕ¨ + d · ϕ˙ + Md .

(8)



526

00025

J. Martin et al.

Fig. 11. Model structure for a finger joint, the valve properties were determined separately.

However, measurements with predefined pressure trajectories and Md = 0 show obvious differences to the simulation (Fig. 13), especially a hysteresis and a timedependent behavior, which can be attributed to the actuator material. Contrary to metals, polymer materials do not show a linear elastic behavior (σ = ε · E). Rather, an asymptotic increase in strain ε(t) at a constant tension σ or an asymptotically decreasing tension σ(t) can be observed at constant strain ε. Such a reversible viscoelastic behavior is typically described in the literature20 with spring-damper models. In the simplest case, this is accomplished with a series of a spring and a second spring with a parallel damper, thus resulting in D(t) =

1 = D0 + c · [1 − e−t/τi ]. E(t)

(9)

This model can also be expanded to a more exact equation (10) using the nonnegative function f (τ ) as a retardation spectrum: ∞ D(t) = D0 + f (τ ) · [1 − e−t/τ ]dτ. (10) 0

An approach allowing to model arbitrary pressures p(t) can be derived using the following assumptions: • the material retardation only affects the pressure-independent part of the actuator torque; • a creeping ratio A with 0 ≤ A ≤ 1 is used to model the pressure influence; • the current creeping ratio results in a displacement of Ma (pa = 0, ϕ) = a0 + a1 ϕ in ϕ. Positive values of A result in reduced actuator reverse forces and therefore increased joint angles; • the maximum retardation-dependent angle ϕmax limits the displacement of ϕ. Hence, we obtain a modified actuator model: Ma (pa (t), ϕ) = a0 + a1 (ϕ + A · ϕmax ) + a2 pa + a3 pa ϕ .

(11)

ϕ

For the integration into the modified actuator model, a material model was designed to describe the hysteresis and the creeping behavior. Figure 12 (left) shows


00025

527

Fig. 12. Graphic representation of the material model.

a

3 2 1 0

50

100 t [s]

150

0

ϕ [˚]

hysteresis

80 70 60 50 40 30 20 10 0

pressure

5

drift

simulation measurement

p [105 Pa]

80 70 60 50 40 30 20 10 0

pressure

p a [10 Pa]

simulation measurement

ϕ [˚]



3 2 1 0

50

100 t [s]

150

0

Fig. 13. Dynamic actuator behavior without material model (left) and with material model (right).

three zones of recovery, constant creeping ratio, and extension separated by p0 (A) ˙ i.e. the rate at which A changes: and p1 (A), the right graph of Fig. 12 denotes A, + A = b1 · (pa − pmin ) for pa ≥ p1 (A), A˙ = 0 for p0 (A) < pa < p1 (A), b0 ˙ for pa ≤ p0 (A) . b2 A + A = b0 · pa b1 ˙ b3 A

(12)

Solving the mathematical description of these graphs in Eq. (12) for A(t=0) = A0 and a pressure step to pa at t ≥ 0, we obtain  b b − 3 ·t − 3 ·t  for pa ≥ p1 (A),  b1 · (pa − pmin ) · 1 − e b1 + A0 · e b1 A(t) = A0 for p0 (A) < pa < p1 (A),  b2 b2  b0 · pa · 1 − e− b0 ·t + A0 · e− b0 ·t for pa ≤ p0 (A) . (13) We obtain a structure identical to Eq. (9), which allows to match the parameters. Using a suitable set of parameters ϕmax , pmin , and b0 , . . . , b3 , we obtain an optimized actuator model behaving as in Fig. 13, right.


528

00025

J. Martin et al.

3.4. Pneumatic control

• adaptation to compensate the non-linear system behavior; • anti-reset-windup integral action and output limiters; • speed up with additional, model-based feed-forward control using the reference variable; • dead-band to reduce valve activity. The controller’s performance with and without adaptation is demonstrated in Fig. 14. Due to its design, the position controller will respond to permanent angular deviations with an actuator pressure rising continuously to the system pressure. This behavior can be considered as undesirable especially for the grasping of delicate objects, as it would result in grasping forces only limited by Ma (ϕ, pmax ). We are therefore evaluating a torque-limited position control with a maximum torque as a second feed-forward reference variable. The actuator model is then used to dynamically limit the maximum actuator pressure depending on the joint position. This feature is discussed in Ref. 21 together with a model-based approach for object contact detection.

ϕ [˚]


The previously described model is the basis for the design of a closed-loop joint position controller. A discrete PID controller implemented in LabView was enhanced with a series of measures:

ϕref (t)

80 70 60 50 40 30

ϕ(t) (with adaptation) ϕ(t) (without adaptation)

0

5

10

15 t [s]

20

25

30

Fig. 14. Comparison of controller behavior with and without adaptation.

4. Conclusion and Outlook We have presented a set of low-cost sensors for the integration into an artificial anthropomorphic hand driven by fluidic actuators. A joint angle sensor based on hall sensors and custom-made rare-earth magnets was evaluated for its output signal properties, linearity, and options for further size reduction. An integration approach for FSR sensors was described considering sensor linearity issues, the detection of low contact forces, and the realization of a soft finger surface especially. Reasons for the use of pressure sensors were given in the context of an actuator model describing the dependencies between actuator pressure, joint angle and the exerted torque. Dynamic actuator characteristics were derived from a theoretical analysis


00025



529

and measurements. A controller design and a number of optimizations were outlined and an outlook on further improvements was given. Future work will cover the port of the current controller algorithm to a microcontroller and the integration of a pressure control loop to replace the analog pressure control valve. Hydraulic joint position and torque control will be developed to reduce the unit’s size and weight.22 Model-based contact detection during joint movements will be evaluated in comparison to direct contact force measurements. Independently, methods for high-level finger coordination and hand control are being evaluated to allow a number of different grasp patterns and the coordination with arm movements.23 Acknowledgment This work was supported in part by the German Research Foundation (DFG) within the Collaborative Research Centre SFB 588 — “Humanoide Roboter — Lernende und kooperierende multimodale Roboter.” References 1. S. Schulz, C. Pylatiuk and G. Bretthauer, A new ultralight anthropomorphic hand, in Proc. 2001 IEEE Int. Conf. Robotics and Automation, Seoul, Korea (IEEE Computer Society Press, 2001), pp. 2437–2441. 2. S. Schulz and G. Bretthauer, A fluidic humanoid robothand, in Proc. 2001 Int. Conf. Humanoid Robots, Waseda University, Tokyo, Japan, November, 2001, pp. 133–139. 3. S. Schulz, C. Pylatiuk, J. Martin and G. Bretthauer, Die Anthropomorphe FZKHand, in Robotik 2002 Leistungsstand — Anwendungen — Visionen — Trends (VDI-Berichte 1679), eds. R. Dillmann, R. D. Schraft and H. W¨ orn (VDI/VDEGesellschaft Mess- und Automatisierungstechnik, 2002), pp. 531–536. 4. J. Martin, R. Keppler, D. Osswald, W. Burger, K. Regenstein, G. Bretthauer, J. Wittenburg, H. W¨ orn, A. Albers and K. Berns, Mechatronische Konzepte zur Verbesserung der Mensch-Maschine-Interaktion, in Tagungsbeitr¨ age der Human Centered Robotic Systems, Karlsruhe, Germany, December, 2002, pp. 65–72. 5. J. Butterfaß, M. Grebenstein, H. Liu and G. Hirzinger, DLR-hand II: Next generation of a dextrous robot hand, in Proc. 2001 IEEE Int. Conf. Robotics and Automation, Seoul, Korea (IEEE Computer Society Press, 2001), pp. 109–114. 6. J. Butterfaß, G. Hirzinger, S. Knoch and H. Liu, DLR’s multisensory hand — Part I: Hard- and soft-ware architecture, in Proc. 1998 IEEE Int. Conf. Robotics and Automation, Leuven, Belgium (IEEE Computer Society Press, 1998), pp. 2081–2086. 7. M. C. Carrozza, S. Micera, B. Massa, M. Zecca, R. Lazzarini, P. Canelli and P. Dario, The development of a novel biomechatronic hand — Ongoing research and preliminary results, in Proc. 2001 IEEE/ASME Int. Conf. Advanced Intelligent Mechatronics (AIM), Como, Italy, July 8–11, 2001, pp. 249–254. 8. M. C. Carrozza, B. Massa, S. Micera, M. Zecca and P. Dario, A wearable artificial hand for prosthetics and humanoid robotics applications, in Proc. 2001 Int. Conf. Humanoid Robots, Waseda University, Tokyo, Japan, November, 2001. 9. J. P. Gazeau, S. Zeghloul, M. Arsicault and J. P. Lallemand, The LMS hand: Force and position control in the aim of the fine manipulation of objects, in Proc. 2001 IEEE Int. Conf. Robotics and Automation, Seoul, Korea (IEEE Computer Society Press, 2001), pp. 2642–2648.



530

00025

J. Martin et al.

10. Th. Fischer and J. Seyfried, The new Karlsruhe Dexterous Hand II, in Int. Symp. Intelligent Robotic Systems, 1997. 11. D. Meeker, Finite element method magnetics (FEMM) user’s manual, Version 3.3, May, 2003. 12. I. Harvey, Artificial evolution: A continuing {SAGA}, in Evolutionary Robotics: From Intelligent Robots to Artificial Life, Procs. 8th Int. Symp. Evolutionary Robotics, Tokyo, Japan, Lecture Notes in Computer Science, Vol. 2217, ed. T. Gomi (SpringerVerlag, Berlin, 2001). 13. M. H. Lee and H. R. Nicholls, Tactile sensing for mechatronics — A state of the art survey, Mechatronics 9, 1–31 (1999). 14. A. Upasani, C. Kapoor and D. Tesar, Survey of available sensor technology for robotic hands, in Proc. 1999 ASME Design Engineering Technical Conf., Las Vegas, Nevada, September 12–15, 1999. DETC99/DAC-8656. 15. P. J. Kyberd and P. H. Chappell, A force sensor for automatic manipulation based on the hall effect, Measuring Sci. Tech. 4, 281–287 (1993). 16. M. R. Cutkosky, J. M. Jourdain and P. K. Wright, Skin materials for robotic fingers, in Proc. 1987 IEEE Int. Conf. Robotics and Automation, Raleigh, USA, (IEEE Computer Society Press, 1987), pp. 1649–1654. 17. R. Z¨ ollner, O. Rogalla and R. Dillmann, Integration of tactile sensors in a programming by demonstration system, in Proc. 2001 IEEE Int. Conf. Robotics and Automation, Seoul, Korea (IEEE Computer Society Press, 2001), pp. 2578–2583. 18. M. Mentzel, F. Hofmann, T. Ebinger, J. G¨ ulke, L. Kinzl and N. J. Wachter, Kraftmessung an der Hand mit einem Sensorhandschuh bei Griffen mit submaximaler und maximaler Kraft, Handchir Mikrochir plast Chir 33(1), 52–57 (2001). 19. A. Kargov, C. Pylatiuk, J. Martin and S. Schulz, A comparison of the grip force distribution in natural hands and in prosthetic hands, in Proc. 2nd World Congress of the Int. Soc. Physical and Rehabilitation Medicine (ISPRM), Prague, Czech Republic, 18–22, May 2003. 20. N. W. Tschoegl, The Phenomenological Theory of Linear Viscoelastic Behavior (Springer-Verlag, Berlin, 1989). 21. S. Beck, R. Mikut, A. Lehmann and G. Bretthauer, Model-based control and object contact detection for a fluidic actuated robotic hand, in Proc. 42nd IEEE Conf. Decision and Control, ed. J. J. Zhu (IEEE Press, 2003), pp. 6369–6374. 22. R. Mikut, A. Lehmann and G. Bretthauer, Fuzzy stability supervision of robot grippers, in IEEE Int. Conf. Fuzzy Systems, Budapest, 2004, accepted paper. 23. D. Osswald, J. Martin, C. Burghart, R. Mikut, H. W¨ orn and G. Bretthauer, Integrating a robot hand into the control system of a humanoid robot, in Proc. 2003 Int. Conf. Humanoid Robots, Karlsruhe und M¨ unchen, Germany (CD-ROM), Oktober, 2003, p. 37.

Jan Martin received his M.Sc. degree in Evolutionary and Adaptive Systems from the University of Sussex, Brighton, UK in 2000. Prior to his graduate studies, he worked in automotive engineering for four years. In 2001, he joined the Institute for Applied Computer Science at the Forschungszentrum Karlsruhe. His main areas in the research on anthropomorphic robot hands are sensors, modeling, control, and movement coordination.


00025


531


Sebastian Beck received his Diplom–Ingenieur degree in Mechanical Engineering from the University of Karlsruhe in 2001. He joined the Institute for Applied Computer Sciences at the Forschungszentrum Karlsruhe in 2001. His research interests are in the areas of decision theory, computational intelligence, decision support systems, and robotics.

Arne Lehmann received his Diplom–Ingenieur degree in Mechanical Engineering from the University of Karlsruhe in 2002. In 2002, he joined the Institute for Applied Computer Sciences at the Forschungszentrum Karlsruhe. His research interests are in the areas of modeling, control, and supervision of robot systems.

Ralf Mikut received his Diplom–Ingenieur degree in Automatic Control at the University of Technology, Dresden in 1994 and his Doktor–Ingenieur degree at the University of Karlsruhe in 1999. He is now a project manager at the Institute for Applied Computer Science at the Forschungszentrum Karlsruhe. His research interests are in the areas of fuzzy systems, computational intelligence, medical diagnosis support systems, automatic control, and robotics. He is elected chairman of the Technical Committee 5.22 “Fuzzy Control” of the German Society of Measurement and Automatic Control (GMA) and Associate Editor of the IEEE Transactions on Fuzzy Systems. Christian Pylatiuk received his M.D. degree in Electro– Neurophysiology from the University of Marburg, Germany in 1997. From 1997 to 1999, he worked as a resident in the orthopedic and surgical departments. Since 1999, he has been a researcher at the Institute for Applied Computer Science and works on the autopropulsive colonoscope and multifunctional prosthetic hand projects. He is responsible for physiological aspects and clinical testing of medical devices and acts as a committed mediator between the parties involved in the project, in particular at the interface of medicine and technology.



532

00025

J. Martin et al.

Stefan Schulz received his Doktor–Ingenieur degree from the University of Karlsruhe in 2004. From 1998 to 1999, he was research associate at the Institute of Applied Computer Science of the Forschungszentrum Karlsruhe and worked on the development of flexible fluidic actuators. Since 1999, he has been leading the fluidic robotics research group, which includes a scientific lab. He is responsible for the conception of artificial hands, the development of the respective components, and the interdisciplinary coordination of the work inside and outside of the working group. His research interests include robotics, micro-mechatronics, and bionics. Georg Bretthauer obtained his Diplom–Ingenieur, Doktor– Ingenieur, and Doktor–Ingenieur habilitatus degrees in Automatic Control at the University of Technology, Dresden in 1970, 1977, and 1983, respectively. He is now Professor of Applied Computer Science and Automatic Control at the University of Karlsruhe and the Head of the Institute of Applied Computer Science at the Forschungszentrum Karlsruhe. His research interests are in the fields of computational intelligence, knowledgebased systems, mechatronics, and automatic control.

SENSORS, IDENTIFICATION, AND LOW LEVEL

SENSORS, IDENTIFICATION, AND LOW LEVEL

Suggest Documents

SENSORS, IDENTIFICATION, AND LOW LEVEL ... - Semantic Scholar

Simple Unsupervised Identification of Low-level ... - CiteSeerX

Oil level sensors

Probe and Sensors Development for Level ...

LFP Cubic Level Sensors - Mysick.com

Dynamical modelling, identification and software sensors ... - CiteSeerX

Low-Level Information and High-Level Perception

Ultrasonic Level Sensors - Omega Engineering

TJE Low - Honeywell Test and Measurement Sensors

Low Cost Accelerometer Sensors - Applications and Challenges

Identification of low-level point radioactive sources using a ... - CiteSeerX

Identification of Low-Level Point Radiation Sources Using a Sensor ...

Identification of novel rapamycin derivatives as low-level impurities in

Identification of novel rapamycin derivatives as low-level ... - Nature

Learning Low-Level Vision

Low-level - UiO

mercedes level sensors - European Braking Systems

Lightweight Liquid Level Sensor - Gill Sensors & Controls

Low numeracy and dyscalculia: identification and intervention

Novel Multimode Interference Liquid Level Sensors

On Whitespace Identification Using Randomly Deployed Sensors

Closed-Loop System Identification with New Sensors

Word level Script Identification from Bangla and

Modelization, Failures Identification and High-level ...