pt .2 01 4 Workspace Augmentation of Spatial 3-DOF Cable Parallel Robots Using Differential Actuation 4- 18 Se Hamed Khakpour and Lionel Birglen, Member, IEEE, pr in to 85 38 0, 88 .3 pp ), 14 20 S O (IR s em st fa pa pe rf ro m th e IE EE /R SJ In te rn at io na lC on fe re n ce on In an s ot ob tR en te llig Cable robots are a special type of parallel robots in which cables are used instead of rigid limbs to manipulate the moving platform (MP) with respect to the base platform (BP) [1]. Using cables in the architecture of these mechanisms brings interesting characteristics. For instance, in addition to the advantages of linkage-driven parallel robots with respect to their serial counterparts (e.g. higher dexterity, payload, and precision), cable-driven designs offer a simpler structure, lighter moving components, low friction (generally), and larger workplaces [2], [3], [4], [5], [6], [7]. Due to these properties and especially the achieved larger workspaces, cable robots are preferred to other robotic manipulators in some applications such as construction sites, shipyards and airplane hangars [8], [9]. Amongst the many practical examples of these robots, one can cite the wellknown NIST Robocrane [10], Cablecam [11], Falcon [12], and Skycam [13]. On the other hand, parallel cable robots suffer from disadvantages such as limited cable tensions, vibration and cable interference, etc [3], [4]. Because of unidirectional nature of the forces in cables, they can only pull never push on the MP. Therefore, to fully constrain this MP, redundancy in the number of cables is a necessity. This means that with a n-degrees-of-freedom (DOF) mechanism, at least m = n + 1 cables are required [2], [14], [15]. As mentioned in the literature, if more cables (i.e., m > n + 1) are used with a proper arrangement in the architecture of the cable-driven mechanism, then, a larger workspace and better performance can be expected [6], [8]. The unilateralism of cables also results in major differences in the analysis of cable-driven and linkage-driven robots d Sy I. INTRODUCTION Pr e (e.g., how to compute the workspaces of these robots [4], [6], [8], [16]). In general, cable robots are classified as either incompletely or fully restrained [4], [17]. In the former, either the MP is suspended from the ceiling (a.k.a. cable suspended robots [4], [18]) or only m ≤ n cables are in tension. While in the latter, regardless of the external wrench, the forceclosure condition is satisfied [4]. Typically, different aspects of cable-driven robots including their kinematics, wrench-closure and wrench-feasible workspaces (WCW & WFW), cable arrangement and force distribution have been analyzed in the literature. For example, Tadokoro et al. [19] optimized the cable distribution of a cable mechanism to improve its workspace. Pusey et al. [16] proposed an incompletely restrained 6-6 cable suspended robot and optimized its workspace by considering the global dexterity of the mechanism. Shiang et al. [9] designed a 3-DOF cable suspended robot and optimized its force distribution. Jiang et al. [20] analyzed the kinematics of cable suspended mechanisms with two to six cables driven by several aerial robots. Yang et al. [6] introduced a 7-DOF cable-driven robotic arm and studied its workspace. Moreover, Bouchard and Gosselin [1] provided a geometrical approach to study the WFW of two to six-DOF cabledriven mechanisms. Gouttefarde and Gosselin [7] proposed several new theorems to determine the WCW of planar cable parallel robots. Mao and Agrawal [21] designed a 5DOF cable robot with adjustable cable attachment points and improved its workspace by optimizing the locations of the cable connection points. Gouttefarde et al. [22] investigated the WFW of a n-DOF cable-driven mechanism by using an interval analysis based approach. Due to the redundancy in the number of cables of these robots, to independently manipulate them redundancy in actuation is also necessary. Therefore, the cost of the mechanism and its control strategy can become quite challenging especially when the number of cables is increased to obtain better workspaces [6], [8]. Thus, it would be very interesting to keep the number of actuators at minimum while the number of the cables is increased. For this, the idea of using cable differentials in the structure of a cable mechanism was presented by the authors and the properties and requirements of such systems to be used with planar cable architectures have been described in [23]. A closely related design to differentials has been proposed in [24], [25] for building cable robots. Although, it was not recognized as a part of a much larger family of architectures based on differentials as demonstrated in [23]. In this paper, the authors focus ,1 Abstract— In this paper, it is proposed to use spatial differentials instead of independently actuated cables to drive cable robots. Spatial cable differentials are constituted of several cables attaching the moving platform to the base but all of these cables are pulled by the same actuator through a differential system. To this aim, cable differentials with both planar and spatial architectures are first described in this work and then, their resultant properties on the force distribution is presented. Next, a special cable differential is selected and used to design the architecture of two incompletely and fully restrained robots. Finally, by comparing the workspaces of these robots with their classically actuated counterparts, the advantage of using differentials on their wrench-closure and wrench-feasible workspaces is illustrated. H. Khakpour and L. Birglen are with the Department of Mechanical Engineering, Ecole Polytechnique de Montreal, Montreal, QC, Canada. Fax number: (514) 340-5867, e-mails: hamed.khakpour@polymtl.ca, lionel.birglen@polymtl.ca. A f s=- f r B' B Pulley attached to a single-point MP Spring Cables S i Cable C D 4 01 Elliptical curve pt .2 fr S i+1 D' Se f f Elliptical curve BP f 4- Actuator Fig. 2: Direction of the resultant force of the two cables of a planar differential when its actuator is locked and the elliptical curve along which the pulley can move. ,1 BP 85 f 18 Resultant force directions pp Pulley attached to a single-point MP MP ), f s=- f r on spatial architecture and investigate the effect of using spatial differentials in 3-DOF spatial cable-driven robots. For this, the properties of planar and spatial differentials are first presented. Then, the designs of two robots with a single-point MP, the first one with three and the second one with four cable differentials are presented. Finally, two types of workspaces of these novel designs are computed and compared with these of fully actuated cable mechanisms. .3 88 0, 38 Fig. 1: Single-point MP actuated by a cable and pulley differential. (IR O S 20 3-elliptical curve f ff f 14 Spring em s Cable Sy f an d BP II. CABLE DIFFERENTIAL SYSTEMS Actuator S1 f st S2 S3 f (a) S3 S2 S1 (b) en tR ob ot s Fig. 3: (a) Single-point MP connected to a planar q = 3 cable differential when the actuator is locked; (b) the routing of the cable. Pr e pr in to fa pa pe rf ro m th e IE EE /R SJ In te rn at io na lC on fe re n ce on In te llig Differentials are used in many different machines to either distribute the input power to two outputs or combine two inputs into a single output. Since each individual differential has 2-DOF, to obtain a larger number of outputs, these differentials are attached together in serial or parallel patterns, or even combination of them [26], [27]. Examples of commonly found differentials are seesaw mechanisms, planetary gear differentials, bevel gearboxes, and tendonpulley systems, all of which have been noticeably successful in driving underactuated robotic hands [28]. A cable differential consists in several cables connecting the MP to the BP while the power of a single actuator is distributed amongst them through a differential mechanism, while in typical cable robots each cable is driven independently. Since, only one actuator is used in the architecture of a typical differential, only a single force can be controlled and applied to the MP of the robot. This property leads to some differences between the analyses of differentially and individually driven cable manipulators. Mainly, instead of a single tension force produced by each individual cable, the resultant force of all cables of each differential and its limits should be considered. The simplest cable differential might be the cable and pulley system presented in Fig. 1 (the spring is used to fully constrain the MP). As can be seen in this figure, the two cables are pulling on both sides of the pulley which is attached to a single-point MP. In an ideal frictionless case and when the radius of the pulley is negligible, the tension forces in the two sides of the cable are equal. Consequently, the resultant force on the MP always lays on the bisector line of both forces and the MP can be constrained only in this particular direction but free to move in the normal direction of the resultant force. This means that if the actuator of this differential is locked the MP can then move on an elliptical curve (compared to a circle in the case of a non-differentially driven design). The direction of this resultant force yields some interesting properties. As illustrated in Fig. 2, when the pulley is at point A (i.e., equidistant of the attachment points Si and Si+1 on the BP), the resultant force intersects the line segment Si Si+1 at its midpoint C. In this configuration, the mechanism is similar to a single cable attached between points A and C. If the pulley moves towards point B, the intersection then lies on point D. This means that the attachment point on the BP of the “virtual” single cable equivalent to the differential system is not stationary but moves along the line Si Si+1 . This phenomenon leads to improvement of the size of the workspaces of cable robots as will be shown. If the planar differential illustrated in Fig. 1 is used in a spatial robot, then the resultant unconstrained elliptical curve of the MP becomes an ellipsoid (made by rotating this curve around the axis defined by Si Si+1 ). In this case, although the MP can move on the surface of the ellipsoid, the intersection of the resultant force still moves along the line segment Si Si+1 . The apparent number of output cables of a differential, q, can be increased to 3, 4 or more (in the previous example, one had q = 2). The arrangement of cables of differentials with q = 2, 3, and 4 cables were synthesized in [23] for the planar case. It should be noted that a q = 2 differential is necessarily planar since it is defined by three points. On S4 f s =- f r S2 z 3 o S1 S7 S2 S5 S6 BP y o Cables x S1 S7 Cables S2 S3 S9 S8 Single-point MP Cables 85 ,1 S12 38 S10 88 Fig. 4: Single-point MP connected to a spatial q = 3 cable differential in three different positions when the actuator is locked. BP Weighted single-point MP g BP (a) (b) pp .3 f S3 S11 0, S9 4- f Midpoint S1 Actuator f S3 S8 z (IR O S 20 14 ), Fig. 5: Schematic of (a) a cable suspended mechanism with three q = 3 spatial differentials; (b) a fully constrained cable mechanism with four q = 3 spatial differentials. en tR ob ot s an d Sy st em s Thus, as mentioned earlier the three output cables of the differential can also be arranged in a 3-D fashion. The schematic of an example of a spatial q = 3 differential is illustrated in Fig. 4. In this case, the routing of the cable though the pulleys on the BP and MP are the same as the schematic presented in Fig. 3b, while the location of point S1 is offset from the previous planar arrangement to obtain a spatial structure. As shown in this figure, the single-point MP is connected to the vertices of a triangle through the three parts of a cable. When the actuator is locked, the MP is free to move on all directions which are perpendicular to the resultant force of the three cable tension forces. In this case, the resultant surface on which the MP can freely move is a 3-ellipsoid. When the MP moves on this surface, the intersection of the resultant force and the BP accordingly moves inside this triangle. As it can be easily seen in this figure, the main difference between a planar and a spatial differential is that in the first case, the aforementioned intersection moves along a line while in the latter it moves on a surface. IE EE /R SJ In te rn at io na lC on fe re n ce on In te llig the other hand, if q > 2 then the differential can be either planar or spatial. As an example, a planar q = 3 differential is depicted in Fig. 3. In this system, a cable is attached to the single-point MP at one end, then after passing around the pulleys respectively at points S2 , S3 and the MP, it is connected to an actuated winch at its other end (for better understanding, the routing of the cable is illustrated in Fig. 3b). Similarly to the previous cases, only the resultant force of all three output cables needs to be considered to compute the WCW and WFW. It should be noted that if the actuator is locked, the direction of this force changes when the MP moves along a 3-ellipse not a simple two focal points ellipse. A 3-ellipse is a planar closed curve consisting of all the points on which if the MP is placed, the sum of its distances from the three attachment points on the BP is constant [29]. Due to the use of a spool in the winch (connected to the actuator) of the cable differential system there is no theoretical limit in the total length of its cable and accommodate any location of the MP. Thus, the aforementioned elliptical curve can have an arbitrary configuration. The location, p(x, y), of the points of this 3ellipse for a known total distance d from the three foci is obtained as [29]: p(x, y) = det A (1) III. PERFORMANCE OF DIFFERENTIALLY DRIVEN SPATIAL CABLE ROBOTS A. Architecture of the differentially driven cable robots Pr e pr in to fa pa pe rf ro m th e where A is an 8 × 8 symmetric matrix in which the diagonal elements are functions of d, x, x1 , x2 , and x3 while the other components are function of y, y1 , y2 , and y3 . Also, S1 (x1 , y1 ), S2 (x2 , y2 ), and S3 (x3 , y3 ) are the coordinates of the attachment points of the cables on the BP in the local coordinate system of the differential. Since this matrix is quite large, it is not presented here. The obtained components of p(x, y) are polynomials of degree eight [29]. The effect of this planar q = 3 differential on the workspace of a planar cable robot (made of three similar differentials) was presented by the authors in [23]. In this paper, spatial differentials are taken into account and their ability to improve the performance of 3-D cable-driven robots is investigated. 4 x 01 S4 2 f ff fr y BP S6 18 1 S5 pt .2 Spring Se Pulley attached to a single-point MP To evaluate the effect of using spatial differentials in the performance of cable robots, two mechanisms are introduced. The schematic of these two designs are presented in Fig. 5. For the sake of simplicity a single-point MP is used in these mechanisms. The first one is a 3-DOF cable suspended robot in which the MP is suspended from the ceiling by three q = 3 spatial differentials. In this robot, the weight of the MP is used to fully constrain the system and all the three differentials are attached to the top of the robot base. The second one is a 3-DOF fully constrained cable robot with a similar MP but four q = 3 spatial differentials. In this case, the MP is assumed to be weightless and thus, in its analysis, only the tensions of the cables and the external wrench exerted to the MP are considered. S'2 z o x In this paper, the effect of using differentials on cable robots is investigated through measuring the sizes of their WCW and WFW. The WCW is a workspace in which the MP can be located while all cables are in tension and their force vectors can fully constrain the motion of the MP. In cable suspended robots, the weight of the MP is used as a passive constraining force and thus, the term static workspace (SW) is used instead of WCW to get rid of any confusion with the usual definition found in the literature. The WFW is a space where the tensions of all the cables are within safe pre-specified limits. To calculate the SW of the robots, the methodology presented in [4] is used. For this, unit vectors along the resultant forces of cables of each differential are first defined as: 01 4 B. Workspace analysis S'1 y BP S'2 S'3 Cables x 38 BP S'4 Weighted single-point MP .3 88 g 85 ,1 Cables 0, S'3 4- 18 Single-point MP Se S'1 y o pt .2 z BP pp (a) (b) ec(3i−2)+ec(3i−1)+ec(3i) er(i) = ec(3i−2)+ec(3i−1)+ec(3i) for i = 1, 2,· · ·, k (IR O S 20 14 ), Fig. 6: Schematic of a fully driven (a) cable suspended robot with three cables and (b) a fully constrained cable robot with four cables. particular pose. If it is larger than the minimal required force, that pose is then inside the WFW of the mechanism. In the presented cable suspended robot, the MP should be heavy enough to bring the origin of the zonotope to its center and consequently maximize the volume of the sphere inside this geometry. In the proposed fully restrained robot, this weight is not useful and could be minimized. Pr e pr in to fa pa pe rf ro m th e IE EE /R SJ In te rn at io na lC on fe re n ce on In Z = α1 ∆f1 ⊕ α2 ∆f2 ⊕ · · · ⊕ αk ∆fk + s X fimin +fg (3) i=1 where αi ∈ [0, 1], s = k × 3, and fg = mg is the weight vector of the MP. Finally, the radius of the largest sphere which can be contained by this zonotope while its center is attached to the origin is the magnitude of the maximum force that can be resisted by the cable robot in all directions in that em st Sy d an s ot ob tR C. Architecture of similar fully actuated cable robots To serve as a basis for comparison between the spatial differentially driven designs and the independently actuated ones, four fully driven cable mechanisms are considered. Two of these mechanisms have exactly the same architectures as the differentially actuated cases of Fig. 5 but all the cables in these designs are assumed to be independently driven. In the other two fully actuated cases, each differential is replaced by a single actuated cable which is connected to the center of the triangles as illustrated in Fig. 6. en te llig where er(i) is the unit vector along the resultant force of the ith differential; ec(3i−2) , ec(3i−1) , and ec(3i) are the unit vectors along the forces of the three output cables of the ith differential; also, k is the total number of differentials. With the first robot, k = 3 and er(4) is along the weight vector of the MP while in the second mechanism, one has k = 4. Next, as described in [4], the cross products cij = er(i) × er(j) for i, j = 1, 2,· · ·, 4 and i < j are calculated. Afterwards, the dot products dijm = eTr(m) cij for m 6= i, j are obtained. If for each cij there is at least one change in the sign of the dijm and this is valid for all cij s the corresponding pose is then included in the SW of the cable robot. To obtain the WFW, a geometrical method similar to the one described in [1] is used. In this method, the tension limits of all cables are assumed to be between fmin and fmax . Next, the tension vectors ∆fc(j) = (fmax −fmin )ec(j) of all cables are obtained. Since the resultant force of each differential should be considered, the vectors ∆fr(i) = ∆fc(3i−2) +∆fc(3i−1) +∆fc(3i) are calculated. These vectors are then considered as line segments and by computing their Minkowski sum [1] an initial zonotope (a type of convex polytope) is generated. Next, taking into account only the lower limit of the cable tensions (namely fmin ) the position of this polytope is modified. The final situation of this polytope can be used to evaluate the WFW of the robot. The mathematical equation for obtaining this zonotope can be formulated as [1]: s (2) IV. IMPLEMENTATION AND RESULTS An algorithm was developed to find the SW and WFW of all these robots. To do this, all the parameters are first chosen (cf. Table I). In table I, a is the length of the edges of the (equilateral) base triangles and the hexagon on the BP (c.f. Fig. 5), h is the distance between the upper and lower parts of the BP (with the fully constrained designs), α is the rotation angle of the equilateral triangles with respect to the hexagon around their shared edges on the top side of the BP; wmp is the weight of the MP with the cable suspended robots, and fW F W is the minimal magnitude of the external force which should be resisted in all directions by the MP. Also, the algorithm searches for the workspaces of the robots in a cylindrical space with a diameter of dc and an height of hc (c.f. Table I). The base surface of this imaginary cylinder is attached to the bottom side of the BP while its axis coincides with the origin of the inertial frame of the robot. With these numerical values, the workspaces of all robots are obtained and illustrated in Figs. 7-12. Also, the ratios TABLE I: Parameters used in the calculations Values 100 N 0.05 1.2 0.3 −20 WFW −40 −60 SW ,1 4- 0 −80 −40 85 40 X −40 0 0 Y −20 20 40 −40 .3 WFW 0, SW 38 20 −20 88 −20 Z 4 0 01 Par. fmax fmin /fmax wmp /fmax fW F W /fmax pt .2 Values 40 cm 60 cm Se Par. dc hc 18 Values 40 cm 80 cm π/4 rad Z Par. a h α pp Fig. 9: SW and WFW of the cable suspended robot driven by nine independent cables. −80 −40 20 14 ), −60 40 −20 O Y 0 Fig. 7: SW and WFW of the cable suspended robots driven by three independent cables. em s −40 −40 WFW s an Z −20 st 40 Sy 20 (IR 0 0 d X S 20 −20 SW ob ot −60 tR 0 en 40 X on ce −60 re n SW on fe −80 −40 40 Y −40 Fig. 10: SW and WFW of the fully restrained cable robot driven by four independent cables. 40 20 0 0 −20 20 −40 0 Y rn at 40 io na X lC −20 −20 20 In −40 0 0 te WFW 20 −20 llig −20 Z −80 −40 −20 Z WFW −40 IE EE /R SJ In te Fig. 8: SW and WFW of the cable suspended robot driven by three q = 3 spatial differentials. Pr e pr in to fa pa pe rf ro m th e of the volumes of the workspaces to the volume of the cylindrical configuration space are presented in Table II. As can be seen in this table, with both suspended and fully constrained cable robots, the SW and WFW of the differentially driven designs are larger than these with the same number of actuators but each only driving a single cable. On the other hand, the workspaces of differentially driven designs are smaller than these of fully driven mechanisms with the same number of cables when they are all actuated individually. This means that using differentials, the size of the workspaces of cable robots can be expected to be improved. Nevertheless, one should note that using differential may also result in some drawbacks in practice. For instance, the friction between the cables and pulleys may lead SW −60 −80 −40 40 20 −20 0 0 −20 20 X 40 −40 Y Fig. 11: SW and WFW of the fully restrained cable robot driven by four q = 3 spatial differentials. to unbalance distribution of tension forces and using more cables can increase the possibility of interference of cables. But, with a proper design, differentials can indeed increase the performance of these robots. The concept of differentially −40 WFW SW −80 −40 ,1 4- 18 −60 85 40 20 −20 \\\\\\\\ \\\\\\\\ \\\\\\\\ Y \\\\\\\\ \\\\\\\\ \\\\\\\\ X \\\\\\\\ \\\\\\\\ 88 40 −40 0, Y −20 20 38 0 0 X 20 14 ), pp .3 Fig. 12: SW and WFW of the fully restrained cable robot driven by twelve independent cables. O (IR s an d Sy st em s W F W/Clr 0.0107 0.0855 0.3487 0.0426 0.1163 0.3472 In en te llig driven cable manipulator can be used in different fields such as manipulation, rehabilitation, cargo, and space robots. tR ob ot SW/Clr 0.4469 0.5263 0.9054 0.1583 0.1987 0.5443 S TABLE II: Comparing the SW and WFW of the cable-driven robots. Type of the cable robot Three cables Suspended Three differentials with: Nine cables four cables Fully restrained four differentials with: twelve cables on V. CONCLUSIONS IE EE /R SJ In te rn at io na lC on fe re n ce In this paper, spatial cable differentials were presented and their properties compared with common independently actuated cables. Their atypical force distribution was first described. Then, two cable robots either suspended or fully restrained with a similar single-point MP were introduced and driven by respectively three and four q = 3 spatial differentials. The SW and WFW of these robots were then compared with four more classically driven robots. The results revealed that by replacing a single cable with a spatial differentials both types of workspaces can be improved. R EFERENCES pa pe rf ro m th e [1] S. Bouchard, C. Gosselin, and B. Moore, “On the ability of a cabledriven robot to generate a prescribed set of wrenches,” J. Mech. Robot., vol. 2, no. 1, p. 011010 (10 pp.), Feb. 2010. [2] S. Perreault and C. M. Gosselin, “Cable-driven parallel mechanismsapplication to a locomotion interface,” J. Mech. Des. (USA), vol. 130, no. 10, pp. 1 023 011–8, Oct. 2008. [3] S. Tadokoro, Y. Murao, M. Hiller, R. Murata, H. Kohkawa, and T. Matsushima, “A motion base with 6-dof by parallel cable drive architecture,” IEEE ASME Trans Mechatron, vol. 7, no. 2, pp. 115– 123, Jun. 2002. [4] G. Yang, W. Lin, M. S. Kurbanhusen, C. B. Pham, and S. H. Yeo, “Kinematic design of a 7-dof cable-driven humanoid arm: A solutionin-nature approach,” in IEEE ASME Int Conf Adv Intellig Mechatron AIM, Monterey, California, USA, Jul. 2005, pp. 444–9. [5] F. Ferlay and F. Gosselin, “A new cable-actuated haptic interface design,” in Haptics: Perception, Devices and Scenarios. 6th International Conference, EuroHaptics, Madrid, Spain, Jun. 2008, pp. 474–83. fa pr in to Pr e 4 01 Se Z −20 [6] G. Yang, C. B. Pham, and S. H. Yeo, “Workspace performance optimization of fully restrained cable-driven parallel manipulators,” in IEEE Int Conf Intell Rob Syst, Beijing, China, Oct. 2006, pp. 85–90. [7] M. Gouttefarde and C. M. Gosselin, “Analysis of the wrench-closure workspace of planar parallel cable-driven mechanisms,” IEEE Trans. Robot. (USA), vol. 22, no. 3, pp. 434–445, Jun. 2006. [8] A. Fattah and S. K. Agrawal, “On the design of cable-suspended planar parallel robots,” Trans. ASME, J. Mech. Des. (USA), vol. 127, no. 5, pp. 1021–8, Sep. 2005. [9] W. J. Shiang, D. Cannon, and J. Gorman, “Optimal force distribution applied to a robotic crane with flexible cables,” in Proc IEEE Int Conf Rob Autom, Francisco, CA, USA, Apr. 2000, pp. 1948–54. [10] J. Albus, R. Bostelman, and N. Dagalakis, “The nist robocrane,” J Rob Syst, vol. 10, no. 5, pp. 709–24, Jul. 1993. [11] J. Rodnunsky, “Cabling system and method for facilitating fluid threedimensional movement of a suspended camera,” US Patent 7 088 071, 2006. [12] S. Kawamura, W. Choe, S. Tanaka, and S. R. Pandian, “Development of an ultrahigh speed robot falcon using wire drive system,” in Proc IEEE Int Conf Rob Autom, Nagoya, Japan, May 1995, pp. 215–20. [13] R. R. Thompson and M. S. Blackstone, “Three-dimensional moving camera assembly with informational cover housing,” US Patent 6 873 355, 2005. [14] S. Kawamura, W. Choe, S. Tanaka, and S. R. Pandian, “Development of an ultrahigh speed robot falcon using wire drive system,” in IEEE International Conference on Robotics and Automation, vol. 1, Nagoya, Japan, May 1995, pp. 215–20. [15] M. Carricato and J. P. Merlet, “Stability analysis of underconstrained cable-driven parallel robots,” IEEE Trans. Robot., vol. 29, no. 1, pp. 288–96, Feb. 2013. [16] J. Pusey, A. Fattah, S. Agrawal, E. Messina, and A. Jacoff, “Design and workspace analysis of a 6-6 cable-suspended parallel robot,” in IEEE Int Conf Intell Rob Syst, Las Vegas, Nevada, USA, Oct. 2003, pp. 2090–5. [17] A. T. Riechel, P. Bosscher, H. Lipkin, and I. Ebert-Uphoff, “Concept paper: Cable-driven robots for use in hazardous environments,” in Conf. Robot. Remote. Syst. Proc., vol. 10, Gainesville, FL, Mar. 2004, pp. 310–6. [18] M. Gouttefarde, J. Collard, N. Riehl, and C. Baradat, “Simplified static analysis of large-dimension parallel cable-driven robots,” in Proc IEEE Int Conf Rob Autom, Saint Paul, Minnesota, May 2012, pp. 2299– 2305. [19] S. Tadokoro, S. Nishioka, T. Kimura, M. Hattori, T. Takamori, and K. Maeda, “On fundamental design of wire configurations of wiredriven parallel manipulators with redundancy,” in Proceedings of the Japan-USA Symposium on Flexible Automation, vol. 1, Boston, MA, USA, Jul. 1996, pp. 151–8. [20] Q. Jiang and V. Kumar, “The inverse kinematics of cooperative transport with multiple aerial robots,” IEEE Trans. Robot., vol. 29, no. 1, pp. 136–145, Feb. 2013. [21] Y. Mao and S. K. Agrawal, “A cable driven upper arm exoskeleton for upper extremity rehabilitation,” in Proc IEEE Int Conf Rob Autom, Shanghai, China, May 2011, pp. 4163–8. [22] M. Gouttefarde, J. P. Merlet, and D. Daney, “Wrench-feasible workspace of parallel cable-driven mechanisms,” in Proc IEEE Int Conf Rob Autom, Rome, Italy, Apr. 2007, pp. 1492–7. [23] H. Khakpour, L. Birglen, and S. A. Tahan, “Synthesis of differentially driven planar cable parallel manipulators,” IEEE Trans. Robot., vol. 30, no. 3, pp. 619 – 630, 2014. [24] H. Liu and C. Gosselin, “A planar closed-loop cable-driven parallel mechanism,” Trans. Can. Soc. Mech. Eng., vol. 33, no. 4, pp. 587–96, 2009. [25] H. Liu, C. Gosselin, and T. Laliberte, “Conceptual design and static analysis of novel planar spring-loaded cable-loop-driven parallel mechanisms,” J. Mech. Robot., vol. 4, no. 2, pp. 021 001 (1–11), May 2012. [26] IFToMM, “Terminology for the theory of machines and mechanisms,” Mech Mach Theory, vol. 26, no. 5, pp. 435–539, 1991. [27] S. Hirose, “Connected differential mechanism and its applications,” in Proceedings of International Conference on Advanced Robotics, Tokyo, Japan, Sep. 1985, pp. 319–25. [28] L. Birglen, T. Laliberte, and C. Gosselin, Underactuated Robotic Hands. Springer-Verlag, 2008. [29] J. Nie, P. Parrilo, and B. Sturmfels, “Semidefinite representation of the k-ellipse,” Algorithms in Algebraic Geometry, vol. 146, pp. 117–32, 2008. pt .2 0