Virtual Whiskers Highly Responsive Robot Collision Avoidance

Transcription

1 2013 IEEE/RSJ Internatonal Conference on Intellgent Robots and Systems (IROS) November 3-7, Tokyo, Japan Vrtual Whskers Hghly Responsve Robot Collson Avodance Thomas Schlegl 1, Torsten Kröger 2, Andre Gaschler 3, Oussama Khatb 2 and Hubert Zangl 1 Abstract All mammals but humans use whskers n order to rapdly acqure nformaton about objects n the vcnty of the head. Collsons of the head and objects can be avoded as the contact pont s moved from the body surface to the whskers. Such a behavor s also hghly desrable durng many robot tasks such as for human-robot nteracton. Usng novel capactve proxmty sensors, robots sense when they approach a human (or an object) and react before they actually collde wth t. We propose a sensor and control concept that mmcs the behavor of whskers by means of capactve sensors. Major advantages are the absence of physcal whskers, the absence of blnd spots and a very short response tme. The sensors are flexble and thn so that they feature skn-lke propertes and can be attached to varous robotc lnk and jont shapes. In comparson to capactve proxmty sensors, the proposed vrtual whskers offer better senstvty towards small conductve as well as non conductve objects. Equpped wth the new proxmty sensors, a seven-jont robot for humanrobot nteracton tasks shows the effcency and responsveness of our concept. I. INTRODUCTION When robots and humans share the same envronment specal precautons are requred to avod njures. More general, anytme robots operate n changng envronments, ther percepton and control systems have to be aware of the envronment and acqure nformaton by numerous sensors. Ths s actually smlar to the way humans and anmals nteract wth ther envronment. Whle vson plays a major role, several other sensng prncples are used complementary. A very successful concept for proxmty detecton n nature are whskers. All mammals but humans use whskers to sense objects n the vcnty of the head [1]. Although whskers requre some mechancal contact they may also be consdered as pretouch sensors, as the contact does not nvolve the actual body surface. Recently, whskers have been proposed for actve sensng n robotcs applcatons [2]. As shown, whskers are not just smple proxmty sensors, they can provde addtonal nformaton about the surface propertes of the object, whch may be used for a coarse classfcaton. We propose a sensor concept that behaves n a smlar way to physcal whskers but uses a capactve sensng technology. Physcal whskers, such as the facal whskers Ths research was partally funded by the Austran Scence Fund (FWF): P21855-N22 1 T. Schlegl and H. Zangl are wth the Insttute of Electrcal Measurement and Measurement Sgnal Processng, Graz Unversty of Technology, Inffeldgasse 23/2, A-8010 Graz, Austra {thomas.schlegl@eee.org, hubert.zangl@tugraz.at}. 2 T. Kröger and O. Khatb are wth the Artfcal Intellgence Laboratory, Stanford Unversty, Stanford, CA 94305, USA {tkr, khatb}@stanford.edu. 3 A. Gaschler s wth fortss GmbH, afflated wth the Technsche Unverstät München, Munch, Germany gaschler@fortss.org. Fg. 1. The presented capactve sensor detects approachng objects at suffcent dstance and measurement rate and allows the robot to avod collsons n real-tme. of rodents have a tapered rod-lke structure. These structures get deformed by objects durng contact. The force caused by the deformaton s then detected by sensors on the surface. In contrast, our vrtual whskers are formed by an electrc feld that gets deformed by an object wthout any physcal contact. The deformaton of the electrc feld s measured by means of electrodes on the surface of the robot. Consequently, we acheve true pretouch sensng wthout any mechancal nteracton. Furthermore, the sensor adds very lttle bulk to the robot arm, consstng of a surface patch of approxmately 350 µm thckness. Furthermore, we acheve short response tmes and a wde range of detectable objects. Whle physcal whskers can provde addtonal nformaton about the surface propertes (e.g. stckness), our vrtual whskers provde addtonal nformaton about the volume propertes (e.g. conductvty) of objects n the vcnty of the sensor. II. RELATED WORK A huge varety of sensor technologes mght be appled for pretouch sensng, but only a few can cope wth the requrements n robotc applcatons. For example, there are lmtatons wth respect to spatal dmensons, weght and adjustablty n postonng. In [3] an optcal proxmty sensor s presented whch s used to mprove graspng tasks. The detecton range s about 30 mm. Although t works for a lot of materals, the sensor sgnal depends on the color and surface of the objects. Thus, as shown n [3], the sensor wll fal for transparent or reflectve objects, for nstance, a contaner made of glass or alumnum, respectvely. Furthermore, t s dffcult to use /13/$ IEEE 5373

2 optcal pretouch systems for short dstances wthout havng blnd spots, due to the lmted observaton angle of each sensor. Magnetc sensors, such as gant magnetoresstance (GMR) sensors, can be used to detect ferromagnetc objects and conductve objects by means of eddy currents. Snce GMR sensors can be made very small, t s easy to mplement them on a robotc hand or arm. In [4] and [5] GMR sensors were used to detect ferromagnetc and conductve objects wthn a detecton range of about 30 mm. A completely dfferent sensor approach s presented n [6]. The so-called Seashell effect uses the resonant frequency of a cavty, where ths resonant frequency changes wth an approachng object. A mcrophone measures the resonant frequency and thus the dstance to an approachng object can be measured (up to approxmately 6 mm). The authors were able to mount the cavty and the mcrophone nto the fngertps of a robotc hand to mprove graspng tasks. Capactve sensng, sometmes also called electrc feld sensng, has a long hstory for proxmty sensng n varous applcatons. In [7] and [8] capactve sensors for collson avodance for predefned and grounded objects are presented. The measurement range s about 80 mm for a 60 mm dameter steel ppe. In [9] a capactve human detecton system s presented. More recently n [10] and [11] capactve sensng was used for pretouch functonalty for graspng applcatons of dfferent objects. All these capactve sensors use ampltude nformaton out of a parwse electrc feld sensng system (.e. dfferental mode, one transmttng electrode and one or more recevng electrodes or vce versa, compare Sec. III-A) at one frequency wth a rather low measurement speed (e.g. 30 Hz n [10], [11]). Thus, these sensors have lmtatons for certan objects. For nstance, the sensors can show nconsstent readngs dependng on the mpedance to ground of the object beng sensed (compare [4] and Sec. IV). The presented system n ths work e.g. s able to work n both measurement modes, has a varable measurement frequency and a hgher measurement speed (compare Table I). As far as the authors know, the presented sensor system s the frst whch acheves the presented measurement speed and measurement nformaton (not only ampltude but also phase nformaton) to be used n such a hghly reactve robot applcaton (compare Table I). Especally the speed and the use of two dfferent capactve sensng modes makes t sutable to use for collson avodance n robot applcatons (compare Sec. III-A). Furthermore, the sensor s not only senstve to humans, wth the utlzaton of phase and ampltude nformaton provdng addtonal capabltes for detectng conductve and delectrc objects (due to dfferent propertes of the objects parastc connecton to ground, compare Fg. 4) An llustraton of the sensor system n comparson wth physcal whskers s shown n Fg. 2. Instead of rod-lke structures the vrtual whskers are represented by an electrc feld around the body. The deformaton of the electrc feld by an approachng object s llustrated n Fg. 3 (obtaned from a Fnte Element Smulaton). Object Body Object Body Fg. 2. Illustraton of whskers. Left: Physcal whskers, deformed by an object. Rght: Vrtual Whskers, represented by an electrc feld, deformed by an object. Fg. 3. A descrptve example on how an approachng object deforms the electrc feld. A. Sensng Hardware III. SYSTEM DESCRIPTION Capactve sensors usually operate n one of two possble modes: The frst mode, usually denoted as dfferental mode, utlzes measurements of the capactance between two conductve areas (called electrodes) by applyng a voltage on one electrode (transmtter) and measurng the dsplacement current on the other electrode (recever) (compare Fg. 4). The second mode, usually denoted as sngle-ended mode, utlzes measurements of the dsplacement current orgnatng from one electrode. As shown n Fg. 4, the dsplacement current corresponds to the capactance between the electrode (transmtter) and the dstant ground, whch s essentally the envronment. The two measurement modes have dfferent propertes (compare Fg. 5). For example, the dfferental mode can provde M N ndependent measurements, where N s the number of transmtters and M s the number of recevers. As an electrode may be both a transmtter and a recever, the number of ndependent measurements s gven by N (N 1) 2. In contrast, the sngle-ended mode can only provde N ndependent measurements. An advantage of the sngle-ended mode s that t usually features a better sgnalto-nose rato for well-grounded objects. It has been demonstrated prevously that the dfferental mode usng more than two electrodes can not only detect the proxmty to an object but also ts orentaton (compare [12]). A related technology s Electrcal Capactance Tomography (ECT), whch s used n ndustral processes to obtan 2D cross sectonal mage of the materal dstrbuton wthn ppes. ECT s essentally an array of capactve sensors wth heavy sgnal processng to calculate an mage of the regon of nterest [13]. Recently n [4] we showed that the ECT approach can be transferred to robot sensng. Wth a 5374

3 below 1 ms). A bref summary of the sensor propertes s gven n Table I. TABLE I PROPERTIES OF THE PRESENTED MEASUREMENT SYSTEM Fg. 4. Sketch of the capactve measurement prncple for one par of electrodes (Elec1 used as transmtter, Elec2 used as recever). The red arrows ndcate the dsplacement currents that are measured n the sngleended mode (extng Elec1) and n the dfferental mode (enterng Elec2), respectvely. In sngle-ended mode, the guard s connected to the exctaton sgnal (.e., actve guardng). R GND, L GND and C GND ndcate an equvalent crcut for the object s parastc connecton to ground. dfferental mode capactve measurement we demonstrated the basc functonalty, but we also found lmtatons. For example, t may not be possble to detect small conductve objects. Ths s due to two competng effects referred to as couplng and sheldng mode [14]. At transtons between these modes, sgnal cancelaton may occur. As the sheldng mode s strongly related to the sngle-ended mode, ths can be used to dentfy the presence of both modes (compare Fg. 5). Thus, we propose to combne sngle-ended and dfferental mode to avod the blnd spots. Sensor thckness 350 µm Exctaton sgnal Snusodal sgnal Frequency Tunable from 10 khz to 1 MHz Measurement rate 1.25 khz (max MHz) compared to e.g. [11]: 30 Hz or to [15]: max. 200 Hz Measurement method Sngle-ended and dfferental mode others (e.g. [11] or [15]): only one mode Sheldng Actve guardng n sngle-ended mode Grounded sheldng for dfferental mode Number of electrodes N = 7 Number of ndependent ( ) measurements 28 = N (N 1) 2 + N for each frequency Fg. 6 gves an overvew of the sensor hardware. The measurement frequency of the snusodal sgnal can be changed between 10 khz and 1 MHz. Although not used n ths work, a varyng measurement frequency wll especally be useful for object classfcaton purposes, due to frequency dependency of the parastc effects as shown n Fg. 4. The generated sgnal can be appled to one or more electrodes and t s also used as an actve guard for the sngle-ended sensng mode. In the dfferental sensng mode the backsde of the sensor nterface (shown n Fg. 11(a)) s connected to ground. Each sensng electrode can be ether connected to a recever amplfer (for dfferental sensng mode) or to one transmtter amplfer (for sngle-ended sensng). (a) Sngle-ended mode (b) Dfferental mode, neghborng electrodes (c) Dfferental mode, electrode 2 to electrode 6 Fg. 5. Senstvty maps wth respect to grounded conductve objects for dfferent sensng modes as obtaned from Fnte Element Smulaton. The colors correspond to the relatve senstvty (red: hgh, blue: low) at certan locatons wthn the regon of nterest. The seven sensor electrodes are located on the left sde (black lnes). (a) In the sngle-ended mode (transmtter measurement) the senstvty ncreases excessvely for objects that are very close to the electrodes. (b) and (c) In the dfferental mode the senstvty map vares dependng on the par of electrodes. Whle the confguraton for Recever 1 has smlar behavor as for a transmtter measurement, the confguraton for Recever 2 suppresses the excessve senstvty for close objects. Consequently, the sngle-ended mode manly responds on the shortest dstance between a conductve object and an electrode whle the dfferental mode offers better detecton capablty for non conductve objects. The sensor proposed n ths work s able to work n both modes (sngle-ended and dfferental mode) and combnes benefts of both measurement modes. Furthermore, t provdes a hgh measurement rate (sample perod T s Fg. 6. Overvew of the developed capactve measurement system comprsng sgnal generaton, transmtter and recever amplfers, IQ-demodulator, ADC, mcrocontroller, and a USB connecton to a host computer. Dashed lnes ndcate dgtal control buses such as SPI or I2C and unbroken lnes ndcate analog sgnals (e.g., snusodal sgnal from sgnal generator to electrodes). The IQ-Demodulator (where I and Q represent the nphase and the quadrature component, respectvely) s used to get ampltude and phase nformaton of the receved and amplfed sgnals (compared to the exctaton sgnal). The outputs of the IQ-Demodulator are dgtzed and collected n a mcrocontroller whch processes the measurement data and has a seral connecton to a host computer to post process the measurement data. Due to the hgh senstvty and fast acquston, the sensor 5375

4 IV. EXPERIMENTS AND RESULTS Ths work focuses on usng the proposed capactve measurement system n a robotc applcaton for hghly reactve proxmty sensng and collson avodance for robot/human nteracton. The frst subsecton presents the results of usng the capactve sensor to avod collsons wth a human. It also presents the benefts for collson avodance wth other objects, when usng both measurement modes (sngle-ended and dfferental mode). Although the presented capactve measurement system s partcularly sutable for sensng human hands, t can also detect other objects, such as conductve or delectrc ones. The second subsecton presents results of measurng dfferent knds of objects. Fg. 7. Overvew of one of the smplest possble robot moton generaton archtectures. The two sets of moton parameters task and react can be changed from one control cycle to another usng the swtchng sgnal σ. The onlne trajectory generaton algorthms [16] of the Reflexxes Moton Lbrary [17] let robots react to the nput sgnals from the proxmty sensor wthn the same control cycle they occur. A. Hghly Reactve Collson Avodance Experments s sutable to be used as a hghly reactve proxmty sensor n robot applcatons (as shown n Sec. IV.) 0.0 s 0.5 s 0.8 s Fg. 8. A robot arm reacts nstantaneously to the measurement data of the presented capactve sensor. Capactve sensng allows the robot to avod collson wth a human hand, mantanng a mnmum dstance of 50 mm. B. Hghly Reactve Robot Moton Generaton and Control The vrtual whskers can be ntegrated n many ways n robot moton control schemes. Ths subsecton descrbes one out of many possbltes. We ntend to keep the control scheme very smple. The dscrete control scheme shown n Fg. 7 works a samplng perod of T cycle. A state of moton at an nstant T s represented by the robots poston P, ts velocty V, and ts acceleraton A. Takng nto account knematc moton constrants B that contan maxmum values for the velocty, acceleraton, and jerk vectors, the onlne trajectory generaton algorthms [16] of the Reflexxes Moton Lbrares [17] compute a tme-optmal, jerk-lmted, and synchronzed trajectory ( that) transfers the robot system from ts current state P, V, A a desred target poston P trgt and velocty V trgt. These algorthms are executed at every control cycle, so that the system can always react nstantaneously ( n a determnstc ) way. The output of the algorthms P +1, V +1, A +1 s forwarded to the underlyng robot moton controller. The underlyng controller can be a poston controller, a trajectory followng controller, an mpedance controller, or any other controller that s capable of followng a trajectory. As long as no object s detected n the proxmty of the vrtual whskers, the task-dependent nput values P trgt,task, V trgt,task, and B trgt,task are used. At the moment an object s detected, the value of the swtchng varable σ changes, and a dfferent set of nput values P trgt,react, V trgt,react, and B trgt,react are used so that the robot can react mmedately and try to avod the potental collson. 0.0 s 0.6 s 1.1 s Fg. 9. If the capactve sensng system measures n sngle-ended mode, the empty plastc case cannot be detected early enough to avod a collson. Thus the arm touches the object before t moves back. 0.0 s 0.6 s 1.0 s Fg. 10. When usng the capactve sensor n dfferental mode, even an empty plastc case can be detected early enough to avod a collson. A seres of experments wth the capactve sensor mounted on a KUKA Lghtweght Robot IV [18] were accomplshed n order to show the capabltes of the capactve measurement system. The KUKA Lghtweght Robot IV was nterfaced through the Fast Research Interface (FRI) [19]. The onlne trajectory generator provdes new pose set ponts for the jont mpedance control nterface of the FRI at a rate of 1 khz. The electroncs crcuts of the capactve sensors are connected through a seral communcaton channel; new sensor sgnals are sampled at a rate of 1.25 khz. Once an object s detected by the vrtual whskers, the robot recols nstantaneously to a safe pose that was ether pre-defned 5376

5 or computed based on the measured sensor sgnals. For descrpton about the responsveness of the control archtecture please refer to [20]. As can be seen n Fg. 11(a), 3 electrodes (.e. 6 ndependent measurements) are used for the collson avodance approach. In the frst experment (shown n Fg. 8), the robot s movng along a certan trajectory. As soon as t senses (.e., the capactve measurement sgnals, I or Q channel, reach a certan threshold) the human hand, t nstantaneously moves nto the other drecton. After the object has left the senstve area, the robot goes on wth ts orgnal trajectory. The second experment shows the lmtatons of a sngleended measurement systems. An empty plastc case s n the way of the robot s trajectory. Snce the measurement sgnal of the sngle-ended mode s too weak, a collson can not be avoded. After the sensor surface touches the plastc case, the threshold s detected and the robot moves n the other drecton. As shown n Fg. 5(c) the dfferental mode s especally useful for sensng delectrc objects. Compared to the second experment the plastc case can be detected n experment 3 (shown n Fg. 10) usng the dfferental sensng mode. In ths work the mentoned threshold was chosen manually to work wth all tested objects. Thus, the robot arm reacts at dfferent dstances for dfferent objects. In future work the authors want to use object classfcaton to adapt the threshold value to partcular object classes. The object classfcaton wll be possble due to the addtonal measurement nformaton (e.g. ampltude and phase, varyng sensng frequency, etc.) from the sensng hardware. B. Sensng Dfferent Objects Fg. 11(a) shows the sensng electrodes and the objects, whch were used to test wth the presented capactve measurement system. In every experment the used object approaches the sensor surface to a dstance of about 5 mm and departs agan. The same sensng electrodes as for the experments n Sec. IV-A are used. Snce the presented system measures the change n ampltude and phase (.e. I and Q channel), two measurements for the sngle-ended mode and four measurements for the dfferental mode (two recevng electrodes) are recorded. The measurement sgnals for an approachng human hand can be seen n Fg. 12. Both sensng modes are able to detect the approach. In the dfferental mode at a certan dstance to the sensor surface (n Fg. 12 around tme 1.5 s) the transton between couplng and sheldng occurs (refer to Sec. III-A and [14]). The hand s stll gettng closer, but due to the ncreasng couplng effect the measured capactance (.e., receved dsplacement current) ntally decreases; t ncreases agan for very short dstances. Fg. 13 shows the approach of a metallc rod wth a dameter of 7 mm (shown n Fg. 11(b)) n two dfferent confguratons. In the frst confguraton the rod s left floatng (not connected to ground). In the second confguraton a cable s used to connect the metallc rod to ground (bypass the parastcs C GND and L GND shown n Fg. 4). As can Fg. 12. Sensor sgnals for a human hand approachng and leavng the sensor surface n both modes, that s, n sngle-ended mode (a) and n dfferental mode (b). The sensors maxmum sgnal occurs at a dstance of 5mm. be seen n Fg. 13, the measurements n the dfferental mode for the rod connected to ground result n a dfferent measurement compared to the one wthout a connecton to ground. The measurements for the sngle-ended mode do not change wth a dfferent connecton to ground. Thus, n combnng both measurement modes, addtonal nformaton about the approachng object and the envronment can be made and conductve objects can be measured more relably. Fg. 13. Sensor sgnals for an approachng metallc rod. For each mode (sngle-ended mode n (a) and (b) and dfferental mode n (c) and (d)) the metallc rod was approached to the sensor surface (mnmum dstance: 5mm) n two confguratons. In the frst confguraton (n the upper fgures) t was not connected to ground whle n the second confguraton the rod was connected to ground. The developed measurement system s also able to detect delectrc objects. The plastc case n Fg. 11(c) can be detected by the measurement system as shown n Fg. 16. Wth the sngle-ended measurement mode, t s very dffcult to measure certan delectrc objects, e.g. objects wth a very small susceptblty χ and small conductvty σ (compare Fg. 16(a)). Wth the addtonal dfferental mode, the plastc case can be detected and a collson wth the robot arm can be avoded as shown n Fg. 10. Other delectrc objects, such as a wood stck shown n Fg. 11(d)), can also be detected by the presented measurement system (shown n Fg. 14). Smlar to the measurement results wth the plastc case n Fg. 16, the dfferental mode s more approprate to measure these knds of objects, due to the hgher measurement sgnal. 5377

6 (a) Sensng electrodes (b) Metal (thn ron rod) (c) Plastc (PVC) case (d) Wood (e) Foam materal (PUR) Fg. 11. Sensng electrodes mounted on the robot arm n (a) and sample objects to test the capactve measurement system n (b)-(e). (b) An ron rod s used to show the propertes for conductve objects (connected and not connected to ground). (c) Delectrc objects such as ths polyvnylchlorde (PVC) case can be detected even f they are small n volume and have a low susceptblty (χ 2). (d) Also a stck made of wood (χ 2) changes the electrc feld and can thus be detected. (e) The foam materal consstng of polyurethane (PUR), whch has a very small electrc susceptblty χ 1 and conductvty σ 0 shows the lmtatons of the sensor as t cannot be used n collson avodance. Fg. 16. Sensor sgnals for a plastc case approachng to about 5 mm and leavng the sensor surface. Wth only the sngle-ended mode n (a) t would be very dffcult to detect the approachng object (scales dffer by factor of 10). Fg. 14. Sensor sgnals for a stck made of wood approachng to about 5 mm and leavng the sensor surface. Fg. 15. Sensor sgnals for a foam materal approachng to about 5 mm and leavng the sensor surface. Snce the foam s electrc susceptblty χ s very low and t s of small volume/densty t has only lttle nfluence on the electrc feld. Thus, t cannot relably be detected by the measurement system. Capactve sensng for delectrc objects depends, besdes other thngs, on the electrc susceptblty (χ) and the volume of the objects of nterest. Thus, t s very dffcult to measure objects wth a very low susceptblty χ (.e., close to the one of ar, χ,ar 0). As can be seen n Fg. 15, an approachng foam cannot be detected n the dfferental measurement mode. Even n the hgh senstvty sngle-ended mode, the foam can only be detected at very close dstances (< 5 mm) from the sensor. Thus, ths mode cannot avod a collson at a reasonable speed. V. CONCLUSION In ths paper, we presented a pretouch sensng method based on capactve measurements for hghly responsve collson avodance n human-robot nteracton. The sensor acts lke vrtual whskers by measurng the deformaton of an electrc feld caused by the approachng objects. Due to the hgh measurement rate and use of two sensng modes (dfferental mode and sngle-ended mode), the measurement system outperforms state of the art capactve proxmty sensors. Thus, t s not only possble to sense conductve objects but also non conductve (delectrc) objects, whch can be small n volume. Furthermore, ths can be done at hgh measurement rates that are hgh enough to avod robot collsons. A seres of experments on a KUKA Lghtweght Robot IV n combnaton wth the onlne trajectory generaton algorthms for the Reflexxes Moton Lbrares proved the proposed hghly responsve collson avodance behavors durng human-robot nteracton tasks. APPENDIX The vdeo attachment shows a seres of experments usng the KUKA Lghtweght Robot IV. The experments nclude several human-robot collson avodance scenaros (ncludng usng human heads) and the use of several materals whle the two dfferent sensng modes are appled. An HQ verson of the vdeo attachment can be found at: ACKNOWLEDGMENT Ths work was performed whle the frst author was vstng the Stanford Robotcs Laboratory at Stanford Unversty, whose hostng s gratefully acknowledged. 5378

7 REFERENCES [1] R. B. Towal, B. W. Qust, V. Gopal, J. H. Solomon, and M. J. Z. Hartmann, The morphology of the rat vbrssal array: A model for quantfyng spatotemporal patterns of whsker-object contact, PLoS Comput Bol, vol. 7, no. 4, p. e , [2] J. Sullvan, B. Mtchnson, M. Pearson, M. Evans, N. Lepora, C. Fox, C. Melhush, and T. Prescott, Tactle dscrmnaton usng actve whsker sensors, Sensors Journal, IEEE, vol. 12, no. 2, pp , [3] K. Hsao, P. Nangeron, M. Huber, A. Saxena, and A. Y. Ng, Reactve graspng usng optcal proxmty sensors, n Robotcs and Automaton, ICRA 09. IEEE Internatonal Conference on, may 2009, pp [4] T. Schlegl, M. Neumayer, S. Mühlbacher-Karrer, and H. Zangl, A pretouch sensng system for a robot grasper usng magnetc and capactve sensors, Instrumentaton and Measurement, IEEE Transactons on, vol. n press. [5] W. Renhart, M. Bellna, C. Magele, and A. Köstnger, Hdden metallc object localzaton by usng gant magnetc resstor sensors, n 14th IGTE Symposum, [6] L.-T. Jang and J. Smth, Seashell effect pretouch sensng for robotc graspng, n Robotcs and Automaton (ICRA), 2012 IEEE Internatonal Conference on, may 2012, pp [7] J. L. Novak and J. Feddema, A capactance-based proxmty sensor for whole arm obstacle avodance, n IEEE Internatonal Conference on Robotcs and Automaton, May 1992, pp Vol.2. [8] J. Feddema and J. Novak, Whole arm obstacle avodance for teleoperated robots, n Robotcs and Automaton, Proceedngs., 1994 IEEE Internatonal Conference on, may 1994, pp vol.4. [9] N. Karlsson, Theory and applcaton of a capactve sensor for safeguardng n ndustry, n Instrumentaton and Measurement Technology Conference, IMTC/94. Conference Proceedngs. 10th Annversary. Advanced Technologes n I amp; M., 1994 IEEE, may 1994, pp vol.2. [10] J. Smth, E. Garca, R. Wstort, and G. Krshnamoorthy, Electrc feld magng pretouch for robotc graspers, n Intellgent Robots and Systems, IROS IEEE/RSJ Internatonal Conference on, Nov. 2, pp [11] R. Wstort and J. R. Smth, Electrc feld servong for robotc manpulaton, n IEEE Internatonal Conference on Intellgent Robots and Systems, September 2008, pp [12] J. Smth, Electrc feld magng, Ph.D. dssertaton, Massachusetts Insttute of Technology, [13] W. Q. Yang and L. Peng, Image reconstructon algorthms for electrcal capactance tomography, Measurement Scence and Technology, vol. 14, no. 1, pp. R1 R13, [14] H. Zangl, T. Bretterkleber, T. Werth, and D. Hammerschmdt, Seat occupancy detecton usng capactve sensng technology, n SAE World Congress, Detrot, MI, USA, SAE paper number , Aprl [15] September [16] T. Kröger, On-Lne Trajectory Generaton n Robotc Systems, 1st ed., ser. Sprnger Tracts n Advanced Robotcs. Berln, Hedelberg, Germany: Sprnger, Jan. 2010, vol. 58. [17], Openng the door to new sensor-based robot applcatons The Reflexxes Moton Lbrares, n Proc. of the IEEE Internatonal Conference on Robotcs and Automaton, Shangha, Chna, May [18] R. Bschoff, J. Kurth, G. Schreber, R. Köppe, A. Albu-Schäffer, A. Beyer, O. Eberger, S. Haddadn, A. Stemmer, G. Grunwald, and G. Hrznger, The KUKA-DLR lghtweght robot arm A new reference platform for robotcs research and manufacturng, n Proc. of the Jont Conference of ISR 2010 (41st Internatonel Symposum on Robotcs) and ROBOTIK 2010 (6th German Conference on Robotcs). Munch, Germany: VDE Verlag, June [19] G. Schreber, A. Stemmer, and R. Bschoff, The fast research nterface for the KUKA lghtweght robot, n Workshop on Innovatve Robot Control Archtectures for Demandng (Research) Applcatons How to Modfy and Enhance Commercal Controllers at the IEEE Internatonal Conference on Robotcs and Automaton, Anchorage, AK, USA, May 2010, pp [20] T. Kröger and J. Padal, Smple and robust vsual servo control of robot arms usng an on-lne trajectory generator, n Proc. of the IEEE Internatonal Conference on Robotcs and Automaton, Sant Paul, MN, USA, May 2012, pp