You are on page 1of 6

Gecko Inspired Micro-Fibrillar Adhesives for Wall Climbing Robots on Micro/Nanoscale Rough Surfaces

Burak Aksak, Michael P. Murphy, and Metin Sitti


Abstract This paper presents the fabrication, characterization and testing of bio-inspired synthetic dry adhesive ber arrays. Fibers were fabricated via micromolding followed by spatula tip formation via dipping. Arrays of bers with diameters between 28 m and 57 m and a height of 114 m were fabricated with high uniformity on 6.25 cm2 areas with up to 95% yield. Adaptation to uneven surfaces was observed with ber elongations over 6 times the original height of the ber. The unstructured sample exhibited 1.6 times as much adhesion as the ber array sample for the at punch indenter. However, brillar samples demonstrated up to 5.3 times as much adhesion as the unstructured sample for the hemispherical indenter. The brillar adhesive sample was implemented on a wall climbing robot which was able to carry itself and climb a distance on a painted wall and wood door.

I. INTRODUCTION Wall climbing robots have the ability to reach areas and perform tasks that ground based and ying robots cannot. A wide variety of climbing robots have been developed in the past decades. In general, climbing robots use one of four types of attachment mechanisms; vacuum suction [1] [7], magnetic attraction [8], gripping with claws or grasping mechanisms [9], [10], or adhesives [11][15]. For very rough surfaces such as brick, gripping has been demonstrated as an effective method [9], and for smooth glass-like surfaces, adhesive attachment mechanisms have been successfully implemented. However, many surfaces that might be desirable for scaling do not fall into either of these two categories (very rough or very smooth), painted structures and walls for example. To investigate the possibilities for climbing such structures we look to nature, where this problem has been solved many times over. Various insects and lizards, notably the Gecko lizard, have specialized footpads which enable them to climb on a wide variety of surfaces of any orientation using dry adhesion. Arrays of micro- or nano-scale hairs create large contact areas with climbing surfaces. Recently there has been signicant research progress in developing synthetic micro- and nanoscale mimics of these adhesive systems. Researchers are characterizing biological samples [16][18] and fabricating synthetic adhesive arrays from polymers [19], [20], carbon nanotubes [21], [22], using methods such as micro/nanomolding [12], [20], [23][28], nano-embossing [29], carbon nanotube growth [21], [22], ber drawing [30], and lithography [19], [31], [32]. In a previous study we fabricated spatula tipped bers and studied their adhesion to hemispherical surfaces [28].
All authors are with the Department of Mechanical Engineering, Carnegie Mellon University, Pittsburgh, Pennsylvania, Sitti@cmu.edu

In this paper we investigate the rough surface adaptation properties of these bers by examining their adhesion against at and curved surfaces and comparing the results with at unstructured control sample. To demonstrate the advantages of spatula tipped bers over the at unstructured sample for rough surface adhesion, we implemented these adhesives as footpads in a wall-climbing robot which climbs on non-smooth surfaces. In this paper, Section II describes the advantages of brillar adhesives. Fabrication processes, experimental apparatus and procedures, and samples are described in Section III. Results are presented and discussed in Section IV. Initial tests and application scenarios are described in Section V. Finally, conclusions and future directions are reported in Section VI. II. F IBRILLAR A DHESIVES One advantage of brillar adhesives over at unstructured adhesives is that each ber deforms independently, which allows them to access deeper recessions to make contact. Even with the reduced total area due to the spaces between the bers, the actual contact area can be greater than that of a at adhesive in contact with a rough surface (Fig. 1). When a at adhesive contacts a rough surface, contact is only made at the highest asperities, and deformation of the bulk layer is relatively small. Because of their structures, brillar adhesives have a much lower effective Youngs modulus, and can deect more while conforming to surface roughness. This allows larger surface roughness asperities to be tolerated as well as some forms of contamination. Details of roughness adaptation can be found in [28], [31], [33]. Although the contact area at each tip can be small, the summation of the contact areas of all of the bers in contact can be quite signicant, particularly if the bers can stretch or deect and remain in contact for large extensions.

Fig. 1. The contact area of a at polymer against a rough surface (a) can be less than the contact area of a brillar adhesive against the same surface (b).

Another advantage of brillar surfaces is their ability to enhance adhesion by contact splitting [34]. If contact is split into many ner independent contacts, adhesive strength

increases due to load sharing. However, adhesive force is directly proportional to both adhesive strength and contact area. To exploit the advantage from brillar adhesives, the enhancement from contact splitting must compensate for the reduction in contact area due to the area missing between the ber tips. III. M ATERIALS AND M ETHODS A. Sample Preparation A two step fabrication method is used to form the ber arrays with spatula tips. Initially, vertical or angled cylindrical bers were created using a fabrication technique detailed in previous work [31]. In the second fabrication step, detailed in previous work [28], a dipping technique was used to form spatula tips on the existing cylindrical bers. The techniques described in [28], [31] were modied for better uniformity and larger areas by carefully controlling the backing layer thickness (to approximately 1.1mm). Using this fabrication method, yield as high as 95% is achievable for ber patches as large as 2.5 cm by 2.5 cm. The dimensions for the resulting bers used in the following experiments are 57 m stem diameter, 114 m tip diameter, 113 m length, and 13 angle with respect to vertical. The center-to-center spacing between the square packed bers was 160 m. In addition to ber array samples, a at unstructured sample was included for comparison.

on an inverted optical microscope (Eclipse TE200; Nikon, Melville, NY, USA) to enable the observation of contact between the indenter and the sample. The indenter was connected to a high resolution load cell (GSO-25; Transducer Techniques Inc. Temecula, CA, USA) which was attached to a high precision stage (MFA-CC; Newport, Irvine, CA, USA). The stage was placed on a two-axis goniometer (GON40-U; Newport, Irvine, CA, USA) to facilitate alignment between the indenter and the sample by observing the contact. The ber sample arrays were placed on the microscope stage bers facing up toward the indenter. The stage, controlled by the custom real-time software, lowers the indenter at a xed speed (1 m/s) until a pre-specied preload (Pp ) is reached. Then the indenter is retracted at the same speed until detachment occurs. The maximum separation force recorded during retraction is called adhesion (Pa ). Slow approach and retraction speeds were chosen in order to minimize viscoelastic effects. By varying the preload, performance curves were created for each sample by plotting the adhesion (Pa ) vs. the preload (Pp ). Two different indenters, a 1 mm diameter silicon disk and a 6 mm diameter glass hemisphere, were used to estimate the adhesion enhancement and degradation of fabricated bers over unstructured samples. The glass hemisphere indenter represents a rough surface which has a well-dened height distribution and a wavelength larger than the size of the bers. Therefore the experiments performed with the hemisphere provide insight into the adhesion performance against a rough surface. On the other hand, the atomically smooth at silicon disk is relevant to the adhesion performance on smooth at surfaces. In addition to the adhesion measurement system, a prole view system [31] was used to qualitatively observe the interaction of ber arrays in contact with a surface.

Fig. 2. Scanning Electron Microscope images of 57 m diameter angled ber array sample illustrating a) High uniformity, and b) spatula tips and a 13 angle with respect to vertical.

B. Apparatus A custom adhesion characterization system was used to perform adhesion measurements. The system was built

Fig. 3. Video still frames of 28 m diameter brillar structures contacting a 6 mm glass hemisphere. a) The two surfaces before contact; b) The surfaces are moved into contact; c) The bers stretch to remain in contact as the surfaces are moved away from each other and shifted laterally; d) Fibers stretch to more than 6 times their initial length.

Fiber arrays were placed into contact with surfaces and

observed under the prole view system. The 6 mm diameter glass sphere used in the adhesion measurements was used as the indenter in these observations. In the prole view system it was observed that the bers stretched up to 6.6 times their initial length before pull-off occurred (Fig. 3d). After pull-off, the bers returned to their initial congurations (Fig. 3a) and were not visibly damaged by the deformation. A video of this interaction is available online [35]. The bers were also stretched laterally (Fig. 3c) to observe behavior under combined shear and normal loading conditions that are common in climbing animals and robots.

a)
Adhesion [ mN ]

100 80 60

120 100 80

60 40 20 1 2 3 4 5 6 7 8 9 10 0

40

20 00

Cycle #

b) 70
60 50

90 80 70

40 30 20 10 0 0 Fiber Array Unstructured Surface

50

40 30 20

Fig. 4. Video still frames of 28 m diameter brillar structures contacting a surface with varying height prole which is overlayed by a white line for clarity. a) The two surfaces before contact, b) The surfaces are moved into full contact where the ber tips touch the recessed areas, c) The surfaces are moved away from each other and the bers stretch to remain in contact. The bers in the recessed areas and on the asperities stretch and remain in contact at the same time, d) the bers in the recessed areas pull off and return to their initial congurations.

10 5 10 15 20 25 0

Preload Force [ mN ]

c) 80
70 60

Fiber Array Unstructured Surface

Another surface that was used to examine the interaction of brillar adhesives with uneven surfaces was a silicon wafer with SU8 photoresist low aspect ratio structures which made uneven steps on the surface of the wafer. The bers were able to attach in the recessed areas and stretch to remain in contact during retraction as seen in Fig. 4c. These observations suggest that these ber arrays exhibit enhanced adhesion against rough surfaces. IV. R ESULTS AND D ISCUSSION The repeatability of adhesion of the ber array was investigated by performing adhesion tests at 8 mN preload for ten cycles on the same area (Fig. 5a). After two cycles the adhesion dropped from approximately 95 mN to 40 mN for the remainder of the experiments. The subsequent data in the at punch adhesion vs. preload gure were obtained after the repeatability tests were preformed without moving the sample. The reduction in adhesion after the rst two measurements may be due to damage to the bers from excessive stretching or surface degradation. After the dropoff in adhesion following the rst two measurements, there appeared to be no signicant degradation. The adhesion vs. preload results for the silicon at punch indenter tests are plotted in Fig. 5b. For both samples the variation of adhesion with increasing preload was minimal. The at sample had a approximately constant adhesion value of 65 mN whereas the ber sample exhibited approximately

Adhesion [ mN ]

50 40 30 20 10 0 0

Preload Force [ mN ]

10

12

14

16

18

Fig. 5. a) Repeatability test data for the 1 mm at punch adhesion tests with a constant 8 mN preload. Adhesion vs. preload data for the 1 mm diameter at punch (b) and 6 mm diameter hemisphere (c) indenters taken at a speed of 1 m/s on arrays of 57 m diameter angled spatula tip bers.

40 mN of adhesion. As discussed in Section II, one possible mechanism for adhesion enhancement is contact splitting. Since the silicon punch is completely at, rough surface adaptation enhancements are not seen and the experimental results reveal the effect of contact splitting only. In this case, the increased adhesion due to contact splitting was not great enough to overcome the reduction in total contact area. The spatula tips cover only 36% of the total area of the punch, the rest of the area is the space between the bers. However, considering only the actual contact area, the brillar sample exhibits higher adhesion per unit area than the control sample, indicating that contact splitting does have

Adhesion [ kPa ]

Adhesion [ mN ]

60

Adhesion [ kPa ]

a signicant effect. In contrast to the at punch indenter experimental results, in the glass hemisphere experiments, the brillar adhesive sample exhibited signicantly higher adhesion than the unstructured surface for every preload tested and the difference grew with increasing preload (Fig. 5c). For low preloads, the brillar sample exhibited 2.5 times (25 mN:10 mN) as much adhesion as the unstructured sample. At the highest preload tested, the brillar sample exhibited 5.3 times (79 mN:15 mN) as much adhesion as the unstructured control sample. This relative change in behavior is attributed to the ability of the bers to stretch over large distances, allowing bers with different stretched lengths to remain in contact simultaneously as seen in Figures 3d and 4c, in other words, roughness adaptation. This evidence of roughness adaptation suggests that the brillar adhesives will exhibit superior performance to the unstructured sample on rough surfaces, as long as the wavelength of the rough surface is larger than the tip size. Considering that the brillar adhesion values in the at punch experiments were fairly close to the unstructured results, the ndings suggest that the brillar sample could be a more versatile attachment pad. For climbing on rougher surfaces, which is signicantly more challenging, the brillar adhesives are far superior to unstructured samples, while sharing similar performance on smooth surfaces. V. A PPLICATIONS

perfectly smooth, so the adhesives will not support as much load on these surfaces. With the brillar adhesive footpads the robot could carry its own weight (78 g) on a vertical painted interior wall (Fig. 6) and a wooden door (Fig. 7), both with visible surface roughness. The surface roughness was high enough to prevent the robot from sticking to the wall at all with at unstructured adhesive pads. Using the brillar footpads, the robot was able to climb a few steps up the vertical wall and door before detaching from the surface (Fig. 7). A video of this climbing trial is available online [36]. In this trial, with freshly fabricated adhesive pads, the robot climbed successfully four steps and then lost contact with the surface. Since the robot climbed four steps and has three footpads per side, this means that while the robot had fresh footpads, it climbed successfully, but when the footpads that had been on the wall previously were carrying the weight of the robot, they were unable to support the load and the robot fell. In subsequent attempts, the robot took successively fewer steps each time before falling, and soon was unable to adhere to the surfaces at all. These observations suggest that contamination of the brillar footpads severely degraded the performance. Contamination was observed by eye and optical microscopy after the tests. Another possible explanation for the decreasing performance of the pads is the degradation of the brillar pads as seen in Figure 5a. When the robot was placed on vertical acrylic, smooth painted steel, and laminated wood surfaces, even with fouled footpads, the brillar adhesives adhered so thoroughly that the actuators could not peel them away to climb.

Fig. 6. Photo of Waalbot climbing a painted interior wall using polyurethane ber arrays as attachment pads.

To demonstrate the effectiveness of the brillar adhesives, they were implemented as footpads on a tetherless wall climbing robot called Waalbot [11], named after the van der Waals forces which it dominantly uses to climb. Waalbot has previously used unstructured elastomer adhesives for attachment and was able to climb smooth surfaces such as acrylic and glass. Extrapolating the results in Figure 5b, two of the robots 19 mm diameter footpads should be able to support 2.89 kg of load in the normal direction if climbing on a perfectly smooth surface. However, in reality, most surfaces are not

Fig. 7. Video frames of Waalbot climbing vertically up a wood door. (a) shows the starting position and (b) shows the nal step before detachment. The overlayed lines illustrate the position of the robot after each of the four steps.

The initial performance of Waalbot with the brillar adhesive footpads suggests that these brillar adhesives show promise as repeatable adhesives for non-smooth surfaces, if the degradation issue can be solved. VI. C ONCLUSIONS A fabrication method was developed and used to form polyurethane adhesive ber arrays with areas of 6.25 cm2

with spatula tips and excellent uniformity. These samples were characterized using a silicon at punch indenter and a glass hemisphere. The results indicate that the brillar samples exhibit up to 5.3 times as much adhesion as the unstructured sample for the hemisphere indenter (smooth curved surface) whereas the unstructured sample yielded higher adhesion for at punch experiments (smooth at surface). The ndings suggest that the brillar adhesives perform better against uneven (low frequency roughness) surfaces than at unstructured adhesive pads. A prole view system was utilized to observe bers conforming to uneven surfaces via their ability to stretch as much as 6 times their initial length and be compressed without losing tip contact. These brillar adhesives were also observed to adhere to recessed regions and asperities of uneven surfaces simultaneously. Fibrillar adhesive samples were implemented as foot pads for a tetherless wall climbing robot which was able to carry its own weight on a vertical wall whose surface was slightly rough (painted) and to which the unstructured samples would not stick. The robot was able to climb a few steps before detaching from the climbing surfaces. We have recently developed several new modications to the tips of the ber arrays including angled spatula tips for directional adhesion and friction (Fig. 8a) and hierarchical bers (Fig. 8b). Future work includes increasing yield for such processes and modeling and characterizing these adhesive systems, and implementing them into the Waalbot footpads. Other future work includes improving the design of Waalbot to suit the performance characteristics of the brillar adhesive footpads rather than the unstructured pads to improve efciency and overall performance. It is evident that research into self-cleaning or contamination resistance is also required. R EFERENCES
[1] S. Hirose, A. Nagakubo, and R. Toyama, Machine that can walk and climb on oors, walls and ceilings, Int. Conf. on Advanced Robotics, vol. 1, pp. 753758, 1991. [2] S. Ryu, J. Park, S. Ryew, and H. Choi, Self-contained wall-climbing robot with closed link mechanism, Int. Conf. on Intelligent Robots and Systems, vol. 2, pp. 839844, 2001. [3] L. Briones, P. Bustamante, and M. Serna, Wall-climbing robot for inspection in nuclear power plants, Int. Conf. on Robotics and Automation, pp. 14091414, 1994. [4] R. Pack, J. Christopher, J.L., and K. Kawamura, A rubbertuator-based structure-climbing inspection robot, IEEE Int. Conf. on Robotics and Automation, pp. 18691874, 1997. [5] B. Luk, A. Collie, V. Piefort, and G. Virk, Robug iii: a tele-operated climbing and walking robot, UKACC Int. Conf. on Control, vol. 1, pp. 347352, 1996. [6] T. Yano, T. Suwa, M. Murakami, and T. Yamamoto, Development of a semi self-contained wall climbing robot with scanning type suction cups, Int. Conf. on Intelligent Robots and Systems, vol. 2, pp. 900 905, 1997. [7] W. Yan, L. Shuliang, X. Dianguo, Z. Yanzheng, S. Hao, and G. Xueshan, Development and application of wall-climbing robots, IEEE Int. Conf. on Robotics and Automation, pp. 12071212, 1999. [8] J. Grieco, M. Prieto, M. Armada, and P. Gonzalez de Santos, A six-legged climbing robot for high payloads, Int. Conf. on Control Applications, pp. 446450, 1998.

Fig. 8. Scanning Electron Micrographs of arrays of 20 m diameter vertical bers with: a) angled spatula tips for directional adhesion and friction; b) 4 m diameter spatula tipped bers integrated into the top of each ber for a hierarchy of compliance and contact splitting.

[9] S. Kim, A. T. Asbeck, M. R. Cutkosky, and W. R. Provancher, Spinybot II: Climbing hard walls with compliant microspines, Int. Conf. on Advanced Robotics, pp. 601606, 2005, 0-7803-9178-0. [10] T. Bretl, S. Rock, and J.-C. Latombe, Motion planning for a threelimbed climbing robot in vertical natural terrain, Int. Conf. on Robotics and Automation, pp. 29472953, 2003. [11] M. P. Murphy and M. Sitti, Waalbot: An agile small-scale wallclimbing robot utilizing dry elastomer adhesives, IEEE/ASME Trans. on Mechatronics, vol. 12, no. 3, pp. 330338, 2007. [12] C. Menon, M. Murphy, and M. Sitti, Gecko inspired surface climbing robots, Proc. of the IEEE Int. Conf. on Robotics and Biomimetics, pp. 431436, 2004. [13] K. Daltorio, S. Gorb, A. Peressadko, A. Horchler, R. Ritzmann, and R. Quinn, A robot that climbs walls using micro-structured polymer feet, in International Conference on Climbing and Walking Robots, 2005, pp. 131138. [14] S. Kim, M. Spenko, S. Trujillo, B. Heyneman, V. Mattoli, and M. Cutkosky, Whole body adhesion: hierarchical, directional and distributed control of adhesive forces for a climbing robot, in IEEE International Conference on Robotics and Automation, 2007, pp. 12681273. [15] O. Unver, A. Uneri, A. Aydemir, and M. Sitti, Geckobot: a gecko inspired climbing robot using elastomer adhesives, in IEEE International Conference on Robotics and Automation, vol. 1, 2006, pp. 23292335. [16] K. Autumn, Y. A. Liang, S. T. Hsieh, W. Zesch, W. P. Chan, T. W. Kenny, R. Fearing, and R. J. Full, Adhesive force of a single gecko foot-hair, Nature, vol. 405, pp. 681685, 2000. [17] K. Autumn, A. Dittmore, D. Santos, M. Spenko, and M. Cutkosky, Frictional adhesion: a new angle on gecko attachment, Journal of

Experimental Biology, vol. 209, pp. 35693579, 2006. [18] K. Autumn, C. Majidi, R. E. Groff, A. Dittmore, and R. Fearing, Effective elastic modulus of isolated gecko setal arrays, Journal of Experimental Biology, vol. 209, no. 18, pp. 35583568, 2006. [Online]. Available: http://jeb.biologists.org/cgi/content/abstract/209/18/3558 [19] A. K. Geim, S. V. Dubonos, I. V. Grigorieva, K. S. Novoselov, A. A. Zhukov, and S. Y. Shapoval, Microfabricated adhesive mimicking gecko foot-hair, Nature Materials, vol. 2, pp. 461463, 1 June 2003. [20] N. J. Glassmaker, A. Jagota, C.-Y. Hui, and J. Kim, Design of biomimetic brillar interfaces: 1. making contact, Journal of The Royal Society, Interface, vol. 1, no. 1, pp. 2333, Nov. 2004. [21] Y. Zhao, T. Tong, L. Delzeit, A. Kashani, M. Meyyappan, and A. Majumdar, Interfacial energy and strength of multiwalled-carbonnanotube-based dry adhesive, Journal of Vacuum Science & Technology B: Microelectronics and Nanometer Structures, vol. 24, pp. 331335, 2006. [22] P. Dickrell, S. Sinnott, D. Hahn, N. Raravikar, L. Schadler, P. Ajayan, and W. Sawyer, Frictional anisotropy of oriented carbon nanotube surfaces, Tribology Letters, vol. 18, no. 1, pp. 5962, 2005. [Online]. Available: http://dx.doi.org/10.1007/s11249-004-1752-0 [23] C. Majidi, R. E. Groff, Y. Maeno, B. Schubert, S. Baek, B. Bush, R. Maboudian, N. Gravish, M. Wilkinson, K. Autumn, and R. S. Fearing, High friction from a stiff polymer using microber arrays, Physical Review Letters, vol. 97, no. 7, p. 076103, 2006. [Online]. Available: http://link.aps.org/abstract/PRL/v97/e076103 [24] M. Sitti and R. Fearing, Synthetic gecko foot-hair micro/nanostructures as dry adhesives, Journal of Adhesion Science and Technology, vol. 17, no. 5, pp. 10551074, May 2003. [25] C. Majidi, R. Groff, and R. Fearing, Clumping and packing of hair arrays manufactured by nanocasting, Proc. of the ASME International Mechanical Engineering Congress and Exposition, pp. 579584, 2004. [26] S. Kim and M. Sitti, Biologically inspired polymer microbers with spatulate tips as repeatable brillar adhesives, Applied Physics Letters, vol. 89, no. 26, p. 261911, 2006. [Online]. Available: http://link.aip.org/link/?APL/89/261911/1

[27] S. Gorb, M. Varenberg, A. Peressadko, and J. Tuma, Biomimetic mushroom-shaped brillar adhesive microstructure, Journal of The Royal Society Interface, vol. 4, pp. 271275, 2007. [28] M. Murphy, B. Aksak, and M. Sitti, Adhesion and anisotropic friction enhancements of angled heterogeneous micro-ber arrays with spherical and spatula tips, Journal of Adhesion Science and Technology, vol. 21, no. 12, pp. 12811296, October 2007. [Online]. Available: http://dx.doi.org/10.1163/156856107782328380 [29] D. S. Kim, H. S. Lee, J. Lee, S. Kim, K.-H. Lee, W. Moon, and T. H. Kwon, Replication of high-aspect-ratio nanopillar array for biomimetic gecko foot-hair prototype by uv nano embossing with anodic aluminum oxide mold, Microsystem Technologies, vol. 13, no. 5, pp. 601606, 2007. [Online]. Available: http://dx.doi.org/10.1007/s00542-006-0220-1 [30] G. L. Spina, C. Stefanini, A. Menciassi, and P. Dario, A novel technological process for fabricating micro-tips for biomimetic adhesion, Journal of Micromechanics and Microengineering, vol. 15, pp. 1576 1587, 2005. [31] B. Aksak, M. Murphy, and M. Sitti, Adhesion of biologically inspired vertical and angled polymer microber arrays, Langmuir, vol. 23, no. 6, pp. 33223332, 2007. [32] N. J. Glassmaker, A. Jagota, C.-Y. Hui, W. L. Noderer, and M. K. Chaudhury, Biologically inspired crack trapping for enhanced adhesion, Proceedings of National Academy of Sciences, vol. 104, pp. 10 78610 791, 2007. [33] C. Greiner, A. delCampo, and E. Arzt, Adhesion of bioinspired micropatterned surfaces: Effects of pillar radius, aspect ratio, and preload, Langmuir, vol. 23, no. 7, pp. 34953502, 2007. [34] E. P. Chang, C. Greiner, E. Arzt, and A. J. Crosby, Designing model systems for enhanced adhesion, Materials Research Society Bulletin, vol. 32, pp. 496543, 2007. [35] [Online]. Available: http://nanolab.me.cmu.edu/projects/geckohair/ [36] [Online]. Available: http://nanolab.me.cmu.edu/projects/waalbots/trileg.html

You might also like