Self-motions of pentapods with linear platform Georg Nawratila, Josef Schichob aInstitute of Discrete Mathematics and Geometry, Vienna University of Technology, Wiedner Hauptstrasse 8-10/104, 1040 Vienna, Austria bJohann Radon Institute for Computational and Applied Mathematics, Austrian Academy of Sciences, Altenberger Strasse 69, 4040 Linz, Austria Abstract We give a full classification of all pentapods with linear platform possessing a self-motion beside the trivial rotation about the platform. Recent research necessitates a contemporary and accurate re-examination of old results on this topic given by Darboux, Mannheim, Duporcq and Bricard, which also takes the coincidence of platform anchor points into account. For our study we use bond theory with respect to a novel kinematic mapping for pentapods with linear platform, beside the method of singular-invariant leg-rearrangements. Based on our results we design pentapods with linear platform, which have a simplified direct kinematics concerning their number of (real) solutions. Keywords: Pentapod, Borel Bricard Problem, Bond Theory, Self-Motion, Direct Kinematic, Kinematic Mapping 1. Introduction The geometry of a pentapod (see Fig. 2a) is given by the five base anchor points Mi with coordinates Mi := (Ai, Bi,Ci)T with respect to the fixed system Σ0 and by the five collinear platform anchor points mi with coordinates mi := (ai, 0, 0)T with respect to the moving system Σ (for i = 1, . . . , 5). All pairs (Mi, mi) of corresponding anchor points are connected by SPS-legs (or alternatively SPU-legs with aligned universal joints; cf. [1, Fig. 1]), where only the prismatic joints are active. If the geometry of the manipulator is given, as well as the lengths of the five pairwise distinct legs, a pentapod has generically mobility 1 according to the formula of Gr¨ubler. In the discussed case of pentapods with linear platform the degree of freedom corresponds to the rotation about the carrier line p of the five platform anchor points. This rotational motion is irrelevant for applications with axial symmetry as e.g. 5-axis milling, spot-welding, laser or water-jet engraving/cutting, spray-based painting, etc. (cf. [2, 3]). Therefore these mechanisms are of great practical interest. In this context configurations should be avoided, where the manipulator gains an additional uncontrollable mobility (beside the rotational motion around p), which is referred as self-motion within this article. Before we give a review on pentapods possessing these special motions in Section 1.2, we repeat a few basics in geometry, which are essential for the understanding of the paper. 1.1. Geometric basics We consider the projective closure of the Euclidean 3-space, which means that we add a point at infinity to each line, which is a so-called ideal point. Moreover two lines are parallel or coincide if and only if they have the same ideal point. The set of ideal points of a pencil of lines (cf. Fig. 1a) form a so-called ideal line of the carrier plane of this pencil. Again two planes are parallel or coincide if and only if they have the ideal line in common. The set of ideal points of a bundle of lines (cf. Fig. 1b) constitute the so-called ideal plane. Summed up we can say that we obtain the projective closure by addition of the ideal plane. Points, lines and planes, which are no ideal elements are called finite. We can also introduce projective point coordinates by homogenizing the coordinates (x, y, z) of a finite point by (1 : x : y : z). The ideal point of a finite line in direction (u, v, w) has the coordinates (0 : u : v : w). Now we can define a regular/singular projectivity (projective mapping) by multiplication of the projective point coordinates with a regular/singular 4 × 4 matrix. If the set of ideal points is mapped onto itself, then the projectivity is called an affinity (affine transformation). arXiv version 2015 arXiv:1407.6126v4 [cs.RO] 16 Oct 2015 (a) (b) gκ Giκ G jκ Gkκ Glκ g Gi G j Gk Gl (c) Figure 1: (a) A pencil of lines is the 1-parametric set of all coplanar lines through a common point, which is the co-called vertex of the pencil. (b) A bundle of lines is the 2-parametric set of all lines through a common point, which is the co-called vertex of the bundle. (c) A regulus of lines. Under a regular projectivity a line g is mapped onto a line gκ due to the linearity of the mapping. The set of lines [Gi, Giκ] with Gi ∈g is called a regulus of lines (cf. Fig. 1c) if g and gκ are skew. In this context it should be noted that the cross-ratio (CR) of four collinear points (cf. [4, page 20]) is invariant under a regular projectivity; i.e. CR(Gi, G j, Gk, Gl) = CR(Giκ, G jκ, Gkκ, Glκ). Conic sections are well-known geometric objects. If the ideal line of the carrier plane of the conic section touches the conic, then it is a parabola. If there are conjugate complex (resp. two real) intersection points, we get an ellipse (resp. a hyperbola). The conjugate complex intersection points of a circle with the ideal line are the co-called cyclic points of the circle’s carrier plane. A further geometric object used within the article at hand is a so-called cubic ellipse. According to [5] a cubic ellipse is a space curve of degree 3, which intersects the ideal plane in one real and two conjugate complex ideal points1. If the latter ones are the cyclic points of a plane (not) orthogonal to the direction of the real ideal point, then the cubic ellipse is called straight cubic circle (skew cubic circle). For more details we refer to [6]. Finally we need the notation of a so-called M¨obius transformation τ of the plane. If we combine the planar Cartesian coordinates (u, v) to a complex number w := u + iv, then τ(w) can be defined as a rational function of the form τ : w 7→z1w + z2 z3w + z4 , (1) with complex numbers z1, . . . , z4 satisfying z1z4 −z2z3 , 0. Moreover it should be noted that τ maps straight lines onto straight lines or circles and that a M¨obius transformation is uniquely defined by three pairwise distinct points w1, w2, w3 and their pairwise distinct images. 1.2. Review, motivation and outline The self-motions of pentapods with linear platform represent interesting solutions to the still unsolved problem posed by the French Academy of Science for the Prix Vaillant of the year 1904, which is also known as Borel-Bricard problem (cf. [7, 8, 9]) and reads as follows: ”Determine and study all displacements of a rigid body in which distinct points of the body move on spherical paths.” For the special case of five collinear points the Borel-Bricard problem seemed to be solved since more than 100 years, due to the following results (cf. [10, page 415]): If five points of a line have spherical trajectories then this property holds for all points of the line. The centers are located on a straight line (cf. Darboux [11, page 222]), a conic section (cf. Mannheim [12, pages 180ff]) or a straight cubic circle (cf. Duporcq [13]; see also Bricard [8, Chapter III]). In a recent publication [14] the authors determined all pentapods with mobility 2, where neither all platform anchor points nor all base anchor points are collinear. As a side result of this study we obtained the following three designs of pentapods with linear platform possessing a self-motion of p: 1Therefore the projection in direction of the real ideal point yields an ellipse. 2 M3 M1 M2 M5 M4 m5 m4 p m3m2 m1 (a) M4 M1 M5 M3 M2 m5 = m3 m4 m1 = m2 h (b) M4 M1 M5 M3 M2 m5 m3 m4 m1 = m2 h (c) M4 M1 M5 M3 M2 m5 m3 m2 m4 m1 h (d) Figure 2: (a) Pentapod with linear platform p (and planar base). (b,c,d) The designs (α, β, γ) are illustrated in a pose of a self-motion. (α) m1 = m2 = m3: The self-motion is obtained if m1 = m2 = m3 is located on the line h spanned by M4 and M5. (β) m1 = m2 and M3, M4, M5 collinear: The self-motion is obtained if m1 = m2 is located on the line h spanned by M3, M4, M5. (γ) M2, M3, M4, M5 collinear: The self-motion is obtained if m1 is located on the line h spanned by M2, M3, M4, M5. In all three cases, which are illustrated in Fig. 2, the following legs can be added without restricting the spherical self-motion: Every point of p can be connected with any point of the line h with exception of the point m1 (= center of spherical motion), which can be linked with any point of the fixed 3-space. This already shows that the cases (α, β, γ) are not covered by the above cited results of Darboux, Mannheim, Duporcq and Bricard, respectively. The reason for this is partially hidden in the formulation of the problem, as 100 years ago they studied the conditions for five points of a line to run on spherical trajectories, which already implies that the collinear points are pairwise distinct. Therefore they missed the cases (α, β), but this still does not explain the absence of case (γ). All in all this shows the need of a contemporary and accurate re-examination of the old results, which also takes the coincidence of platform anchor points into account. This is done within the paper at hand, which is structured as follows: In Section 1.3 we give a short review on architecturally singular pentapods as they imply further solutions to our problem. Based on the method of singular-invariant leg-replacements we determine in Section 2 all non-architecturally singular pentapods with linear platform and planar base possessing self-motions. The same method is used in Section 3 to distinguish five different types of pentapods with linear platform and non-planar base. In Section 4 we introduce the theory of bonds with respect to a novel kinematic mapping for pentapods with linear platform. This theory is used in Section 5 for the determination of all non-architecturally singular pentapods with linear platform and non-planar base possessing self-motions. Finally we use the presented results/methods to design pentapods with a linear platform, which have a simplified direct kinematics with respect to the number of (real) solutions (cf. Section 6). 1.3. Architecturally singular pentapods A pentapod is called architecturally singular if in any pose of the platform the rank of its Jacobian matrix is less than five. This condition also has a line-geometric interpretation as the Jacobian is composed of the Pl¨ucker coordi- nates of the five carrier lines of the legs (cf. [4]). All architecturally singular pentapods are known (cf. [15, Theorem 3] under consideration of [16]) as well as their properties of being redundant and invariant under projectivities of the platform and the base. If we take additionally the collinearity of the platform into account (i.e. m1, . . . , m5 collinear) we end up with the following list: Corollary 1. If a pentapod with linear platform is architecturally singular, then it has one of the following designs 2: 2After a possible necessary renumbering of anchor points. 3 1. M1 = M2 = M3: The first three legs belong to a pencil of lines. 2. m1 = m2 = m3 and M1, M2, M3 are collinear: The first three legs belong to a pencil of lines. 3. M1, . . . , M4 are collinear and the following cross-ratio condition holds: CR(m1, m2, m3, m4) = CR(M1, M2, M3, M4). (2) The first four lines belong to a regulus of lines. In the special case m1 = m2 and M3 = M4 the regulus splits up into two pencils of lines. 4. m1 = m2 = m3 = m4: The first four legs belong to a bundle of lines. 5. M1, . . . , M5 are collinear. 6. m1 = m2 = m3 and M4 = M5. 7. m1 = m2 and m4 = m5. Moreover M1, M2, M5 are collinear and M3, M4, M5 are collinear. 8. m1, . . . , m5 are pairwise distinct, M1, . . . , M5 are coplanar and no three of them are collinear. Moreover there is a projective correspondence between the points mi and Mi for i = 1, . . . , 5. For more details on this most complicated case we refer to [15, Item 10 of Theorem 3] and [17]. 9. m4 = m5 holds and M1, . . . , M5 are coplanar, where M1, M2, M3 are collinear. Moreover the following cross- ratio condition holds: CR(m1, m2, m3, m4 = m5) = CR(M1, M2, M3, M), (3) with M denoting the intersection point of [M4, M5] and the carrier line of M1, M2, M3. In the cases (5-9) the five legs belong to a so-called congruence of lines (cf. [4, Section 3.2.1]). Due to the above mentioned redundancy all these nine cases imply solutions to our problem, as they have a self- motion in each pose of p (over C). Moreover, the architecturally singular cases 2,4,6,7,9 are also not covered by the old results due to the coincidence of platform anchor points. The remaining cases are discussed in more detail: Ad 3: All lines of the regulus can be added without restricting the self-motion. Therefore this case corresponds to the result of Darboux [11, page 222]. Ad 8: The projective correspondence can be extended to all points of the linear platform p and therefore they are mapped onto a conic determined by M1, . . . , M5. Now all legs connecting corresponding anchor points can be attached without changing the self-motion. This equals the solution given by Mannheim [12, pages 180ff]. Ad 1: This case can be interpreted as a special case of Mannheim’s solution, as following legs can be added without restricting the self-motion: Every point of p can be connected with M1 = M2 = M3 with exception of the point mi which can be linked with any point of the line [M1, Mi] for i = 4, 5. Therefore the conic splits up into the two lines [M1, M4] and [M1, M5]. Ad 5: This trivial case can also be seen as a special case of Mannheim’s solution, as the conic degenerates into the double counted carrier line of M1, . . . , M5. As all architecturally singular pentapods with linear platform are already known, we can restrict our study done in the remainder of the article to non-architecturally singular manipulators. Moreover, as the designs (α, β, γ) are not architecturally singular, we can make the following three additional assumptions in order to exclude these already known cases: (i) No three platform anchor points coincide. (ii) If two platform anchor points coincide, the remaining three base anchor points are not collinear. 4 (iii) No four base anchor points are collinear. Definition 1. We define by P the set of all non-architecturally singular pentapods with a linear platform, which fulfill the assumptions (i,ii,iii). We split the determination of all elements of P with self-motions in two parts with respect to the criterion if the base anchor points are coplanar (= planar pentapod; e.g. Fig. 2a,d) or not (= non-planar pentapod; e.g. Fig. 2b,c). 2. Planar pentapods of P with self-motions Within this section we prove the following theorem: Theorem 1. A planar pentapod of P has a self-motion only in the following case: There exists an orthogonal- projection πε of the base plane ε and an orthogonal-projection πp of p in a way that the projected point sets are congruent. In this case p can perform a circular translation. For the proof of this theorem the following preparatory work has to be done: Lemma 1. The anchor points of a planar pentapod with a linear platform, which fulfills the assumptions (i,ii,iii), can always be relabeled in a way that the following conditions hold: M1 , M2, M1, M2, M3 not collinear, M1, M2, M4 not collinear, m3 , m4. Proof: If two platform anchor points coincide we denote them with m4 = m5. Then due to assumption (ii) M1, M2, M3 are not collinear. Due two assumption (iii) one of the remaining two base points is not on the line spanned by M1 and M2. We denote this point by M4 and we are done. Now we discuss the case where all five platform anchor points are pairwise distinct: ⋆If two base points coincide3 then we denote them with M3 = M4. Due to assumption (iii) there are at least two further base points which span together with M3 = M4 a plane. We denote these points by M1 and M2, respectively. ⋆If no base points coincide, but three of them are collinear, then we denote them by M1, M2 and M5. Due to assumption (iii) we are done. ⋆If no three base points are collinear, we can label the points arbitrarily. □ Moreover we can choose the moving frame Σ in a way that m1 equals its origin. The fixed frame Σ0 is selected in a way that M1 equals the origin, M2 is located on the x-axis and the remaining points belong to the xy-plane. Due to Lemma 1 we can assume w.l.o.g. that A2B3B4(a3 −a4) , 0 holds. The proof of Theorem 1 is based on the following result obtained by Borras et al. [17]: A leg of a given planar pentapod with linear platform is replaced by a leg with platform anchor point (a, 0, 0) and base anchor point (A, B, 0) fulfilling Eq. (6) of [17], which reads as follows under consideration of our special choice of coordinate systems Σ and Σ0: [(D2, D3, 0) + a(D4, D5, D1)] A B 1 = 0 (4) with D1 := det(A, B, aA, aB), D2 := −det(a, B, aA, aB), D3 := det(a, A, aA, aB), D4 := −det(a, A, B, aB), D5 := det(a, A, B, aA), (5) 3No three base points can coincide as it yields a contradiction to assumption (iii). 5 and a := a2 a3 a4 a5 , A := A2 A3 A4 A5 , B := 0 B3 B4 B5 , aA := a2A2 a3A3 a4A4 a5A5 , aB := 0 a3B3 a4B4 a5B5 , (6) then the resulting planar pentapod with linear platform has the same singularity set (and direct kinematics solution) if it is not architecturally singular. In the following we study Eq. (4) in more detail: Lemma 2. A planar pentapod with linear platform fulfilling the assumptions (i,ii,iii) is architecturally singular if and only if D1 = D4 = 0 holds (with respect to Σ and Σ0). Proof: Due to Lemma 1 we can assume w.l.o.g. that the following determinant A2 0 0 A3 B3 a3B3 A4 B4 a4B4 = A2B3B4(a3 −a4) (7) is different from zero. Therefore D1 = D4 = 0 implies rk(a, A, B, aA, aB) < 5 which characterizes architecturally singularity (cf. [16, 17, 18]). □ Based on this lemma we can prove the next one, which reads as follows: Lemma 3. For a planar pentapod of P the condition D2 = D3 = 0 cannot hold (with respect to Σ and Σ0). Proof: The proof is done by contradiction as follows: We show that for a planar pentapod with linear platform the condition D2 = D3 = 0 (with respect to Σ and Σ0 ⇒A2B3B4(a3 −a4) , 0) implies either an architecturally singular design (⇔D1 = D4 = 0; cf. Lemma 2) or a contradiction to the assumptions (i,ii,iii). It can easily be seen that D2 = 0 is fulfilled for a2 = 0 (⇔m1 = m2). Now D3 = 0 simplifies to: −a3a4a5A2(A4B5 −A5B4 −A3B5 + A5B3 + A3B4 −A4B3). (8) a3a4a5 = 0 contradicts assumption (i), A2 cannot vanish, and the last factor implies the collinearity of M3, M4, M5, which contradicts assumption (ii). Therefore we can assume for the remaining discussion that a2 , 0 holds. We distinguish the following cases: 1. a5 , 0: Under this assumption we can solve D2 = 0 for A5. Then the numerator of D3 factors into a3a4(A2B3 −A2B4 + A3B4 −A4B3)F[20], (9) where the number in the brackets gives the number of terms. As F is also a factor of D1 and D4, the condition F = 0 implies an architecturally singular design. Moreover ai = 0 (⇔m1 = mi) implies the collinearity of M2, Mj, M5 for pairwise distinct i, j ∈{3, 4}. Therefore a3a4 = 0 contradicts assumption (ii). The vanishing of the third factor of Eq. (9) implies the collinearity of M2, . . . , M5, which contradicts assumption (iii). 2. a5 = 0 (⇔m1 = m5): Now D2 and D3 factors into: −a2a3a4B5(A2B3 −A2B4 + A3B4 −A4B3), a2a3a4A5(A2B3 −A2B4 + A3B4 −A4B3). (10) a2a3a4 = 0 contradicts assumption (i) and the last factor implies the collinearity of M2, M3, M4, which contradicts assumption (ii). Therefore A5 = B5 = 0 has to hold, but in this case the first and the fifth leg coincide. This closes the proof of Lemma 3. □ 6 M3 M1 M2 V M5 M4 m4 = m5 p m2 = m3 m1 (a) M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 1 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 2 = M◦ 3 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 M◦ 4 = M◦ 5 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 1 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 2 = m◦ 3 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 m◦ 4 = m◦ 5 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 S 2 (b) M3 M1 M2 M5 M4 m4 = m5 p m2 = m3 m1 (c) M′ 4 = M′ 5 M′ 2 = M′ 3 M′ 1 m′ 4 = m′ 5 m′ 2 = m′ 3 m′ 1 (d) Figure 3: (a) V is a finite point. (b) Spherical image of the pentapod given in (a) with respect to the unit sphere S 2 centered in V. (c) V is an ideal point. (d) Orthogonal projection of the pentapod given in (c) onto a plane orthogonal to V. Due to this lemma Eq. (4) determines for all planar pentapods of P a bijection between points of p and lines in the base plane.4 Due to the linear relation the lines generate a pencil with vertex V, which can also be an ideal point (⇒ parallel line pencil). According to [19] we are now able to perform a series of leg-replacements in a way that we end up with a non-architecturally singular pentapod of the following type (see Fig. 3a,c): m2 = m3, m4 = m5, M2, M3, V are collinear, M4, M5, V are collinear. (11) Based on this preparatory work we can prove Theorem 1 as follows: Proof of Theorem 1: We distinguish the following two cases: 1. V is a finite point: In this case the motion of p can only be spherical with center V. Therefore we can also consider the spherical 3-legged manipulator, which we obtain by projecting the pentapod onto the unit sphere S 2 centered in V. Therefore the spherical 3-legged manipulator has to have a self-motion. Note that the projected base points M◦ i as well as the projected platform points m◦ i are located on great circles of S 2. It is well known (cf. [20, Lemma 2] and [21, Theorem 5]) that the 3-legged spherical manipulators illustrated in (see Fig. 3b) can only have self-motions if two platform or base anchor points coincide. Now this is only possible if the line p contains V. In this case the platform of the spherical manipulator collapse into a point and we only get the trivial rotation about the line p as uncontrollable motion while p itself remains fix. Therefore this case does not yield a solution. 2. V is an ideal point: Now the motion of p can only be a planar one orthogonal to the direction of V. Therefore the corresponding planar 3-legged manipulator, which is obtained by an orthogonal projection of the pentapod onto a plane orthogonal to V, also has to have a self-motion. Note that the projected base points M′ i as well as the projected platform points m′ i are collinear (see Fig. 3d). According to [20, 21] this planar 3-legged manipulator can only have a self-motion in one of the following two cases: (a) Two platform or base anchor points coincide: This is only possible if the line p contains V. Analogous considerations as in the spherical case show that we do not get a solution. (b) The platform and the base are congruent and all legs have equal lengths: In this case the planar 3-legged manipulator has a circular translation. This already implies the solution given in Theorem 1. □ Remark 1. Note that the case given in Theorem 1 is also not covered by the more than 100 year old results of Darboux, Mannheim, Duporcq and Bricard, respectively, even though no platform anchor points have to coincide. Therefore our study reveals a further lost case beside design (γ). 4For D2 = D3 = 0 Eq. (4) would factor into a(D4A + D5B + D1) = 0, which does not imply such a bijection. 7 But this case is not novel, as it is already contained within the more general characterization given in Theo- rem 25 of [21], which reads as follows: A pentapod has a translational self-motion if and only if the platform can be rotated about the center m1 = M1 into a pose, where the vectors −−−−→ Mimi for i = 2, . . . , 5 fulfill the condition rk(−−−−→ M2m2, . . . , −−−−→ M5m5) ≤1. Therefore a pentapod with a linear platform and a translational self-motion has to have a planar base. ⋄ 3. Types of non-planar pentapods Based on the idea of singular-invariant leg-replacements, which can also be extended to the non-planar case (cf. [2]), one can distinguish different types introduced in this section. Lemma 4. The anchor points of a non-planar pentapod with a linear platform can always be relabeled in a way that the following conditions hold: 1. M1, . . . , M4 span a tetrahedron. 2. If M1, . . . , M5 are pairwise distinct, then M1, Mi, M5 are not collinear for all i ∈{2, 3, 4}. If two base anchor points coincide, then M1 is none of them. As the proof is trivial it is left to the reader and we proceed with the coordinatization used within this section: Again we choose the origin of the moving system Σ in m1. The fixed frame Σ0 is selected in a way that M1 equals the origin, M2 belongs to the x-axis and M3 is located in the xy-plane. All in all this yields A2B3C4 , 0 and M5 , o (zero vector). Moreover we introduce the following notation: Di jk denotes the determinant of the 4 × 7 matrix (a, A, B, C, aA, aB, aC) with C = (0, 0,C4,C5)T, aC = (0, 0, a4C4, a5C5)T (12) after removing the i-th, j-th and k-th column. Now we can state the following lemma: Lemma 5. For a non-planar pentapod of P the condition D167 = D157 = D156 = 0 cannot hold (with respect to Σ and Σ0). Proof: As M2, M3, M4 are linearly independent there exist a unique triple (λ2, λ3, λ4) with λ2M2+λ3M3+λ4M4 = M5. Due to M5 , o we have (λ2, λ3, λ4) , (0, 0, 0). Now it can easily be seen that the following equivalences hold: D167 = 0 ⇔ 4 X i=2 λiaiAi = a5A5, D157 = 0 ⇔ 4 X i=2 λiaiBi = a5B5, D156 = 0 ⇔ 4 X i=2 λiaiCi = a5C5. (13) Therefore D167 = D157 = D156 = 0 implies λ2a2M2 + λ3a3M3 + λ4a4M4 = a5M5. As a5 cannot equal zero6 we can divide both sides by a5, which shows that the following relation has to hold: (λ2, λ3, λ4) = a2 a5 λ2, a3 a5 λ3, a4 a5 λ4 ! . (14) In order to get no contradiction with assumption (i) and M5 , o the implied three equations only have the following solution: ai = a5 and λj = λk = 0 with pairwise distinct i, j, k ∈{2, 3, 4}. For λi , 1 the points M1, Mi, M5 are collinear, which contradicts Lemma 4 and for λi = 1 the i-th leg and the fifth leg coincide; a contradiction. □ In the following we distinguish two cases with respect to the criterion whether D567 vanishes or not. This subdivision was also used by Bricard in [8, Items 12 and 13 of Chapter III] as D567 = 0 is equivalent to the following geometric condition, which we call the affine relation (AR): (AR) There exists a singular affinity κ with Mi 7→mi for i = 1, . . . , 5. 5This theorem is originally stated for hexapods but it also holds for pentapods, as its proof is also valid for 5-legged manipulators. 6For a5 = 0 we get m1 = m5 and therefore a2a3a4 , 0 has to hold, as otherwise we get a contradiction to assumption (i). 8 3.1. D567 , 0 Under this assumption we can use the following result of Borras and Thomas [2]: A leg of a given non-planar pentapod with linear platform is replaced by a leg with platform anchor point (a, 0, 0) and base anchor point (A, B,C) fulfilling Eq. (7) of [2], which reads as follows within our notation: D267 −aD567 −D367 D467 D257 −D357 −aD567 D457 D256 −D356 D456 −aD567 A B C = a D167 D157 D156 (15) then the resulting pentapod has the same singularity set (and direct kinematics solution) if it is not architecturally singular. Solving Eq. (15) with Crammer’s rule yields: A = d1(a) d0(a), B = d2(a) d0(a), C = d3(a) d0(a). (16) Due to the assumption D567 , 0, the polynomial d0 is cubic in the unknown a. The other polynomials d1, d2, d3 are of degree 3 or less in a, but due to Lemma 5 one of them has to be cubic, which shows the following result: Corollary 2. The locus of base anchor points of singular-invariant leg-replacements of a non-planar pentapod of P, which does not fulfill the affine relation (AR), is a cubic space curve. According to Borras and Thomas [2] we can distinguish different types with respect to the number of roots of d0 = 0, for which the system Eq. (15) is consistent. This yields the following classification: Theorem 2. A non-planar pentapod of P, which does not fulfill the affine relation (AR), belongs to one of the following four types: The cubic of Corollary 2: Type 1 is irreducible: There is a bijection σ between p and this space curve s (see Fig. 4a). Type 2 splits up into an irreducible conic q, located in the finite plane ε and a finite line g1 < ε, which intersects q in the point Q: There is a bijection σ between p \ {P1} and q \ {Q}. Moreover the finite point P1 is mapped to g1 (see Fig. 4b). Type 3 splits up into the finite lines l and the finite skew lines g1, g2, which intersects l in the point L1 and L2, respec- tively: There is a bijection σ between p \ {P1, P2} and l \ {L1, L2}. Moreover the finite point Pi is mapped to gi for i = 1, 2 (see Fig. 4c). Type 4 splits up into the finite lines g1, g2, g3, which are not coplanar but intersect each other in the finite point V: All points of p \ {P1, P2, P3} are mapped to V. Moreover the finite point Pi is mapped to gi for i = 1, 2, 3 (see Fig. 4d). Note that in Type i we have 4 −i points W1σ−1, . . . , W4−iσ−1 on p (counted with algebraic multiplicity) for i = 1, . . . , 4, which are mapped by σ to ideal points W1, . . . , W4−i of the base. These points of p have the special property that their trajectory is in a plane orthogonal to the respective ideal point. Therefore each of these point pairs determines a so-called Darboux condition (cf. [8, Item 6 of Chapter II] and [22, Section 4.1]). Any other finite point of p determines a so-called sphere condition; i.e. it is located on a sphere centered in the corresponding finite base anchor point. The points Pi have the special property of possessing circular trajectories, i.e. their path is planar and spherical at the same time. Note that the ideal point U of p is in all four cases mapped by σ onto a finite point Uσ. Therefore this point pair determines a so-called Mannheim condition, which is the inverse of the Darboux condition; i.e. a plane of the moving system orthogonal to p slides through a finite point of the base. Remark 2. The above given correspondence between points on p and points on the base can also be seen as the correspondence of point paths of p and the centers of their osculating spheres. It is an old result of Sch¨onflies [23] that this correspondence is cubic. ⋄ 9 W3σ−1 W2σ−1 W1σ−1 p U W1 W2 W3 Uσ s (a) W2σ−1 P1 W1σ−1 p U g1 q Uσ W1 W2 Q ε (b) P2 W1σ−1 P1 p U L1 g1 L2 W1 Uσ g2 l (c) P3 P2 P1 g3 g2 p U V = Uσ g1 (d) Figure 4: Sketch of the types of non-planar pentapods of P, which do not fulfill the affine relation: (a) Type 1. (b) Type 2. (c) Type 3. (d) Type 4. Pentapods fulfilling the affine relation belong to Type 5. 3.2. D567 = 0 Theorem 3. For a non-planar pentapod of P, which fulfills the affine relation (AR), the ideal point U of p is related to an ideal element of the base within the correspondence implied by singular-invariant leg-replacements. Pentapods with this property belong to Type 5. Proof: Due to Lemma 5 not all three determinants D156, D157, D167 can be equal to zero. Therefore we can assume e.g. that D167 , 07 holds. Then similar considerations to those given in [2] yield the following correspondence: D267 −D367 D467 D126 −aD156 −D136 D146 + aD167 D127 −aD157 aD167 −D137 D147 A B C = a D167 0 0 . (17) Introducing homogeneous coordinates to the first of these three equations show that the ideal point U of p is mapped to an ideal element (ideal point, ideal line or the complete ideal plane) of the base. □ Lemma 6. If a pentapod of Type 5 has a self-motion, then the ideal element has to be a point W. Moreover the self-motion is a Sch¨onflies motion with axis direction W. Proof: We prove by contradiction that U cannot be mapped to more than one ideal point W of the base: Assume that U can be connected with two distinct ideal points W1 and W2 of the base. Then these special two ”legs” correspond to two angle conditions (cf. [8, Item 7 of Chapter II] and [22, Section 4.1]); i.e. the angle enclosed by U and Wi (i = 1, 2) has to be constant during the self-motion. As this already fixes the orientation of p, the pentapod can only have a translational self-motion, which implies the coplanarity of the base (cf. Remark 1); a contradiction. As the angle enclosed by U and W has to be constant, the self-motion of p can only be a Sch¨onflies motion with axis direction W. □ In the following we analyze the Types 1-5 separately with respect to the existence of self-motions. Type 4 can be discussed similar to the planar case, as singular-invariant leg-replacements can be used to get the same pentapod 7For D156 , 0 or D157 , 0 the argumentation can be done analogously. 10 illustrated in Fig. 3a with the sole difference that the three lines through the finite point V are not coplanar. Under consideration of [21, Theorem 6] an analogous argumentation as in item 1 of the proof of Theorem 1 can be done for Type 4, which shows the following result: Theorem 4. A pentapod of Type 4 cannot possess a self-motion. For the discussion of the remaining four types we apply the theory of bonds for pentapods with linear platform, which is the content of the next section. 4. Bond theory for pentapods with linear platform Readers, who are not familiar with the following terms of algebraic geometry (ideal, variety, Gr¨obner base, radi- cal), are recommended to look up these basics in the articles [24] and [25], before studying this section. It was shown by Husty [26] that the sphere condition equals a homogeneous quadratic equations in the Study parameters (e0 : e1 : e2 : e3 : f0 : f1 : f2 : f3). For our choice of the moving frame Σ the sphere condition Λi simplifies to: Λi : (a2 i + A2 i + B2 i + C2 i −R2 i )(e2 0 + e2 1 + e2 2 + e2 3) −2aiAi(e2 0 + e2 1 −e2 2 −e2 3) −4aiBi(e0e3 + e1e2) + 4aiCi(e0e2 −e1e3) −4ai(e0 f1 −e1 f0 −e2 f3 + e3 f2) + 4Ai(e0 f1 −e1 f0 + e2 f3 −e3 f2) + 4Bi(e0 f2 −e1 f3 −e2 f0 + e3 f1) + 4Ci(e0 f3 + e1 f2 −e2 f1 −e3 f0) + 4( f 2 0 + f 2 1 + f 2 2 + f 2 3 ) = 0, (18) where Ri denotes the radius of the sphere centered in Mi on which mi is located. Now, all real points of the Study parameter space P7 (7-dimensional projective space), which are located on the so- called Study quadric Ψ : P3 i=0 ei fi = 0, correspond to an Euclidean displacement, with exception of the 3-dimensional subspace e0 = e1 = e2 = e3 = 0, as its points cannot fulfill the condition e2 0 + e2 1 + e2 2 + e2 3 , 0. The translation vector t := (t1, t2, t3)T and the rotation matrix R of the corresponding Euclidean displacement mi 7→Rmi + t are given by: t1 = −2(e0 f1 −e1 f0 + e2 f3 −e3 f2), t2 = −2(e0 f2 −e2 f0 + e3 f1 −e1 f3), t3 = −2(e0 f3 −e3 f0 + e1 f2 −e2 f1), and R = e2 0 + e2 1 −e2 2 −e2 3 2(e1e2 −e0e3) 2(e1e3 + e0e2) 2(e1e2 + e0e3) e2 0 −e2 1 + e2 2 −e2 3 2(e2e3 −e0e1) 2(e1e3 −e0e2) 2(e2e3 + e0e1) e2 0 −e2 1 −e2 2 + e2 3 , (19) if the normalizing condition e2 0 + e2 1 + e2 2 + e2 3 = 1 is fulfilled. Now the solution for the direct kinematics of a pentapod with linear platform can be written as the algebraic variety of the ideal spanned by Ψ, Λ1, . . . , Λ5, e2 0 + e2 1 + e2 2 + e2 3 = 1. In the general case the dimension of this variety equals 1, as we also obtain the 1-dimensional rotation about the line p. In the following we present another kinematic mapping, where we get rid of this redundancy. 4.1. Kinematic mapping for pentapods with linear platform With respect to the following nine homogeneous motion parameters (n0 : x0 : . . . : x3 : y0 : . . . : y3) with n0 := f 2 0 + f 2 1 + f 2 2 + f 2 3 and x0 := 2(e2 0 + e2 1 + e2 2 + e2 3), y0 := 4(−e0 f1 + e1 f0 + e2 f3 −e3 f2), x1 := 2(−e2 0 −e2 1 + e2 2 + e2 3), y1 := 4(e0 f1 −e1 f0 + e2 f3 −e3 f2), x2 := −4(e0e3 + e1e2), y2 := 4(e0 f2 −e1 f3 −e2 f0 + e3 f1), x3 := 4(e0e2 −e1e3) y3 := 4(e0 f3 + e1 f2 −e2 f1 −e3 f0), the sphere condition of Eq. (18) is linear; i.e. Λi : 1 2(a2 i + A2 i + B2 i + C2 i −R2 i )x0 + aiAix1 + aiBix2 + aiCix3 + aiy0 + Aiy1 + Biy2 + Ciy3 + 4n0 = 0. (20) 11 For x0 = 1 the Euclidean displacement of mi is given by: ai 0 0 7→ −aix1 −y1 −aix2 −y2 −aix3 −y3 . (21) Note that (n0 : x0 : . . . : x3 : y0 : . . . : y3) can be interpreted as a point in the 8-dimensional projective space P8. In order to determine the image of the kinematic mapping, we compute a Gr¨obner bases of the ideal generated by n0 −( f 2 0 + f 2 1 + f 2 2 + f 2 3 ), . . . , y3 −4(e0 f3 + e1 f2 −e2 f1 −e3 f0) and e0 f0 + e1 f1 + e2 f2 + e3 f3 eliminating the Study parameters e0, . . . , f3. It contains three quadric elements in the remaining variables, namely: Φ1 : x2 1 + x2 2 + x2 3 −x2 0 = 0, Φ2 : y2 1 + y2 2 + y2 3 −8x0n0 = 0, Φ3 : x1y1 + x2y2 + x3y3 −x0y0 = 0. (22) Because the degree of the image variety (which is computed from the Gr¨obner bases) is equal to 8, the three quadrics generate the ideal of the image variety I. The image itself consists of all real points in the zero set with x0 , 0. Note that I is of dimension 5 instead of 6, as we removed the rotations around the line p, which do not change the spherical condition. Definition 2. The intersection of I with the five hyperplanes of Eq. (20) is the complex configuration set C of the pentapod. Its real points are called real configurations. Lemma 7. A generic pentapod with linear platform has eight solutions for the direct kinematics over C. Proof: For the solution of the direct kinematics we have to intersect the image variety I with the five hyperplanes of P8 given by Eq. (20). In the generic case the five hyperplanes have a linear 3-space L in common, which intersects I in a finite number of points (= complex configuration set C) whose cardinality equals the degree of I. □ Remark 3. Lemma 7 fits with the results obtained in [27], where the number of eight solutions for the direct kine- matics problem over C was given for planar pentapods with linear platform. Due to Lemma 7 this number also holds for the non-planar case. ⋄ 4.2. Bonds Let us assume that the pentapod under consideration has a 1-dimensional self-motion. In this case the complex configuration set C of Definition 2 is a curve. The points on this configuration curve, which do not correspond to Euclidean displacements as they fulfill x0 = 0, are the so-called bonds. This gives a rough idea of bonds, which is developed in detail within this section. The intersection of the 5-fold I with the hyperplane x0 = 0 yields the so-called boundary of I. A Gr¨obner bases computation of the ideal generated by the generators of the ideal of I and by x0 reveals that some perfect squares, for instance (x2y3 −x3y2)2, are contained in this ideal. Therefore the radical contains the elements Γ4, Γ5, Γ6 below. Its zero set is a 4-fold of degree 4, hence a variety of minimal degree (degree = codimension + 1). It is given by the following set of equations: Γ1 : x2 1 + x2 2 + x2 3 = 0, Γ2 : y2 1 + y2 2 + y2 3 = 0, Γ3 : x1y1 + x2y2 + x3y3 = 0, (23) Γ4 : x1y2 −x2y1 = 0, Γ5 : x1y3 −x3y1 = 0, Γ6 : x2y3 −x3y2 = 0, (24) and therefore it is independent of y0 and n0. Definition 3. The intersection of C with the boundary of I is the set B of bonds. Due to x0 = 0 the set B of bonds is independent of the leg lengths Ri (cf. Eq. (20)), and therefore B only depends on the geometry of the pentapod with linear platform. Remark 4. Bonds of pentapods where already introduced in [21] and [28]. In [21] they were defined with respect to the Study parameters and in [28] with respect to a special compactification of SE(3), which can be seen as a generalization of the method presented above. But both approaches are not suited for the study of pentapods with linear platform, due to the rotational redundancy about p. ⋄ 12 Type Necessary Condition 1 Necessary Condition 2 Sufficiency & Leg Parameters 1 Theorem 6 Theorem 9 Theorem 13 2 Theorem 7 Theorem 10 Theorem 14 3 Theorem 8 5 Theorem 5 Theorem 11 Theorem 12 Table 1: Overview of the Theorems, where the first and second necessary condition for mobility arising from bond theory are applied to the remaining types (for Type 4 see Theorem 4) of non-planar pentapods of P. Moreover the theorems are listed, where the sufficiency of the obtained necessary conditions is proven. Within these theorems also the leg parameters for a self-motion are given. In the following we state two necessary conditions for the existence of self-motions in terms of bonds: Assume that a pentapod with linear platform has a 1-dimensional configuration set C; i.e. there exists a configura- tion curve c on the 5-fold I. Therefore the corresponding bond set B contains at least one bond β (up to conjugation). Therefore the existence of a bond is the first necessary condition. As we have mobility 1, the pentapod with linear platform fulfills the necessary condition of being infinitesimal movable in each pose of the 1-dimensional motion of p. This is equivalent with the existence of a 1-dimensional tangent space in each point of the configuration curve c. As β ∈c holds, this implies a second necessary condition for mobility 1. Remark 5. Therefore the 3-space L has to intersect the 5-fold I in the bond β at least of multiplicity 2. As this also holds for the conjugate of β, only one pair of conjugate complex bonds can exist (due to Bezout’s theorem). Moreover if β (and its conjugate) is a singular point of I then the second necessary condition is trivially fulfilled. ⋄ 5. Non-planar pentapods of P with self-motions The following determination of all non-planar pentapods of P with self-motions is based on the two necessary conditions given in Section 4.2. For an overview of the workflow/results structured by types please see Table 1. 5.1. First necessary condition In this section we only apply the first necessary condition, namely the existence of a bond. Theorem 5. The base anchor points of possible leg-replacements of a pentapod of Type 5 with self-motions have to be located on an irreducible cubic ellipse s∗on a cylinder of revolution. Proof: Given is a pentapod of Type 5. W.l.o.g. we can assume that M1, . . . , M4 span a tetrahedron. Due to the properties of Type 5 (cf. Lemma 6) we can replace the sphere condition implied by the fifth leg by an angle condition; i.e. the angle ϕ enclosed by the ideal point m5 of p and the ideal point M5 of the base is constant. Now we can choose the fixed frame Σ0 that M1 is the origin, M5 is the ideal point of the x-axis, and M2 is located in the xy-plane. The moving frame Σ is chosen in a way that m1 is its origin. With respect to these coordinate systems the angle condition reads as follows: ∢5 : x1 + wx0 = 0 (25) where w denotes arccos (ϕ). The equations of Λi for i = 1, 2, 3, 4 are given in Eq. (20) under consideration of A1 = B1 = C1 = C2 = a1 = 0. We set x0 = 0 and start the computation of the bonds: As M1, . . . , M4 are non-planar (⇔K , 0 with K = A2(B3C4 −B4C3) + B2(A4C3 −A3C4) (26) holds) we can solve Λ1, . . . , Λ4, ∢5 for n0, x1, y1, y2, y3 w.l.o.g.. As we get x1 = 0, the condition Γ1 implies x2 = ±ix3. W.l.o.g. we only discuss the upper branch as the lower one can be done analogously. Now the numerator of Γ3 factors into x3F3[26] and the numerator of Γ4 splits up into x3F4[10]. Therefore we have to distinguish two cases: 13 1. x3 = 0: Now the numerator of Γ2 factors into y2 0F2[41], where F2 is quadratic with respect to a4. The discriminant of F2 with respect to a4 equals: −K2 h (a2A3 −a3A2)2 + (a2B3 −a3B2)2 + a2 2C2 3 i . (27) This expression can never be greater than zero. It is equal to zero if the following three conditions are fulfilled: a2A3 −a3A2 = 0, a2B3 −a3B2 = 0, a2C3 = 0. (28) As a2 = a3 = 0 contradicts assumption (i) we have to discuss the following cases: (a) a2 = 0: This implies A2 = B2 = 0 and therefore the first and second leg coincide; a contradiction. (b) a3 = 0, a2 , 0: We get A3 = B3 = C3 = 0 and therefore the first and third leg coincide; a contradiction. (c) a2a3 , 0: We get C3 = 0, A3 = a3A2/a2 and B3 = a3B2/a2 but this implies K = 0; a contradiction. As a consequence the case x3 = 0 does not imply any solution to our problem. 2. x3 , 0: In this case F3[26] = 0 and F4[10] = 0 have to hold. Computing the resultant of these two expressions with respect to y0 yields −x3K(Gr + iGc) with Gr = [a4B3(a2 −a3) + a3B4(a4 −a2)] B2 + a2(a3 −a4)(B3B4 −C3C4), Gc = −[a4C3(a2 −a3) + a3C4(a4 −a2)] B2 −a2(a3 −a4)(B3C4 + B4C3). (29) These two expressions do not depend on the Ai coordinates. We denote the orthogonal projection of the base anchor point Mi onto the yz-plane of the fixed frame Σ0 by M′ i for i = 1, . . . , 4. In the following we want to show that M′ 1, . . . , M′ 4 has to be pairwise distinct as well as m1, . . . , m4. The proof is done by contradiction, where the following cases have to be discussed: (a) Two platform anchor points coincide: W.l.o.g. we can set m1 = m2; i.e. a2 = 0. Then Gr and Gc simplify to Gr = −a3a4B2(B3 −B4), Gc = a3a4B2(C3 −C4). (30) As a3a4 = 0 contradicts assumption (i), and M′ 3 = M′ 4 (⇒M3, M4, M5 collinear) implies a contradiction to assumption (ii) we remain with the case M′ 1 = M′ 2: In this case m1 = m2 is located on a circle in a plane orthogonal to the axis of the Sch¨onflies motion. As the point m1 = m2 (and therefore the complete line p) cannot be translated in direction of M5, the problem reduces to a planar one (projection to the yz-plane of Σ0). As a circular translation of the resulting planar manipulator is not possible (as otherwise the base has to be planar; cf. Remark 1), anchor points have to coincide (cf. item 2 in the proof of Theorem 1). i. If further base anchor points coincide beside M′ 1 = M′ 2 we get again a contradiction to the assumption M1, . . . , M4 are non-planar. ii. Further platform anchor points (beside m′ 1 = m′ 2) can only coincide without contradicting assumption (i) or (ii) if p is parallel to the axis of the Sch¨onflies motion. In this case the platform of the planar manipulator collapse into a point and we only get the trivial rotation about the line p as uncontrollable motion while p itself remains fix. (b) No platform anchor points coincide and two projected base anchor points coincide: W.l.o.g. we can assume that M′ 1 = M′ 2 holds; i.e. B2 = 0. Then Gr and Gc simplify to Gr = a2(a3 −a4)(B3B4 −C3C4), Gc = −a2(a3 −a4)(B3C4 + B4C3), (31) which cannot vanish without contradiction. 14 As a consequence M′ 1, . . . , M′ 4 have to be pairwise distinct, as well as m1, . . . , m4. Therefore there exists a uniquely defined M¨obius transformation τ with mi 7→M′ i for i = 1, 2, 3 (cf. Section 1.1); i.e. Bi + iCi = τ(ai) holds with τ of Eq. (1) and z1 = a2 −a3 B2 −B3 −iC3 , z2 = 0, z3 = a2B3 −B2a3 + ia2C3 (B2 −B3 −iC3)(B3 + iC3)B2 , z4 = a2a3B2 B3 + iC3 . (32) Now it can easily be verified that the conditions Gr = 0 and Gc = 0 of Eq. (29) imply that τ also maps m4 onto M′ 4. As M1, . . . , M4 span a tetrahedron, M′ 1, . . . , M′ 4 has to be located on a circle. Therefore the base anchor points of possible leg-replacements have to belong to the cylinder of revolution Θ through M1, . . . , M4 with generators in direction of M5. In the following we study the possible leg-replacements for this case in more detail: By homogenizing the matrix of Eq. (12) it is not difficult to see that the corresponding matrix reads as follows: a2 A2 B2 0 a2A2 a2B2 0 a3 A3 B3 C3 a3A3 a3B3 a3C3 a4 A4 B4 C4 a4A4 a4B4 a4C4 0 0 0 0 1 0 0 . (33) As M1, . . . , M4 are not coplanar, we have D167 , 0. But due to D156 = 0 and D157 = 0 Eq. (17) simplifies to D267 −D367 D467 D126 −D136 D146 + aD167 D127 aD167 −D137 D147 A B C = a D167 0 0 . (34) Solving this system of linear equations yields a solution of the form given in Eq. (16), but now d0(a) is a quadratic expression in a and d1(a) a cubic one. Therefore the base anchor points belong to a cubic curve s∗, which has to be located on Θ. Moreover s∗cannot split up into three generators as M′ 1, . . . , M′ 4 are pairwise distinct. The cubic s∗can also not split up into a conic q and a generator g1 for the following reason: This case equals Type 2 where P1 is not a finite point but the ideal point U of p. Therefore Mi has to be located on q \ {Q}, as otherwise mi equals U, which does not yield a sphere condition. As this has to hold for i = 1, . . . , 4 we get a contradiction to the non-planarity assumption. Therefore the cubic curve has to be an irreducible cubic curve s∗on a cylinder of revolution. From the following theorem of projective geometry it is clear that the ideal plane cannot be an osculating plane of s∗or be tangent to it: The osculating plane in a point X of a cubic equals the tangent plane to the cone of chords with respect X along the tangent of X (which is a generator of the cone of chords). Therefore the remaining two intersection points of s∗with the plane at infinity are conjugate complex. □ Theorem 6. The irreducible cubic s of a pentapod of Type 1 with self-motions has to be a cubic ellipse located on a cylinder of revolution. Proof: s has at least one real intersection point with the ideal plane, which is denoted by M4. As the ideal point m5 of p is mapped to a finite point M5 of the base, the point M4σ−1 has to be a finite point of p, which is denoted by m4. Therefore this point pair (M4, m4) determines a Darboux condition Ω4 and the point pair (M5, m5) a Mannheim condition Π5. The remaining three point pairs (Mi, mi) imply sphere conditions Λi for i = 1, 2, 3. Moreover we choose the fixed frame Σ0 that M1 is the origin, M4 the ideal point of the x-axis and M2 is located in the xy-plane. Moreover we can define the moving frame Σ in a way that m1 is its origin. With respect to these coordinate systems our conditions can be written as: Ω4 : p4x0 + a4x1 + y1 = 0, Π5 : p5x0 + A5x1 + B5x2 + C5x3 + y0 = 0, (35) where (p4, 0, 0)T are the coordinates of the intersection point of the Darboux plane and the x-axis of Σ0, and (p5, 0, 0)T are the coordinates of the intersection point of the Mannheim plane and the x-axis of Σ. The equations of Λi for i = 1, 2, 3 are given in Eq. (20) under consideration of A1 = B1 = C1 = C2 = a1 = 0. 15 We set x0 = 0 and start the computation of the bonds: Due to the properties of Type 1 no four base points can be coplanar (⇒B2C3C5 , 0) and no two platform anchor points can coincide. Under these assumptions we can solve the equations Λ1, Λ2, Λ3, Ω4, Π5 for y0, y1, y2, y3, n0. Now the numerator of Γ4 factors into x1F with F := (a2A2 −a4A2 −a2A5)x1 + (a2B2 −a4B2 −a2B5)x2 −a2C5x3. (36) Therefore we have to distinguish two cases: 1. x1 , 0: In this case F = 0 has to hold, which can be solve w.l.o.g. for x3. Then the numerator of Γ6 factors into x2G with G := (A2x1 + B2x2)(a2 −a4)(a3C3 −a3C5 −a4C3)+(A3x1 + B3x2)(a3 −a4)a2C5 −(A5x1 + B5x2)(a3 −a4)a2C3. (37) We distinguish two cases: (a) x2 , 0: In this case G = 0 has to hold. We define H := A2(a2 −a4)(a3C3 −a3C5 −a4C3) + A3(a3 −a4)a2C5 − A5(a3 −a4)a2C3 and discuss the following two cases: i. H , 0: Under this assumption we can solve G = 0 for x1. Then the remaining equations only imply one condition which is quadratic with respect to B3. The discriminant of this condition with respect to B3 equals: −H2 h (a2A2 −a4A2 −a2A5)2 + (a2B2 −a4B2 −a2B5)2 + a2 2C2 5 i . (38) Therefore B3 cannot be real; a contradiction. ii. H = 0: We can solve H = 0 for A3 w.l.o.g.. Then we can solve the G = 0 for B3 w.l.o.g., which already yields the contradiction, as now the points M1, M2, M3, M5 are coplanar. (b) x2 = 0: Now the numerator of Γ1 factors into x2 1 h A2 2(a2 −a4)2 −2a2A2A5(a2 −a4) + (A2 5 + C2 5)a2 2 i . (39) The discriminant with respect to A2 equals −C2 5 and therefore we get a contradiction. 2. x1 = 0: From Γ1 we get x2 = ±x3i. In the following we only discuss the case x2 = x3i, as the other one can be done analogously. Now the numerator of Γ6 factors into x2 3(Gr + iGc) with Gr := (B2B3 −B3B5 + C3C5)a2 −B2(B3 −B5)a3, Gc := B2(C3 −C5)a3 −(B2C3 −B3C5 −B5C3)a2. (40) This are the corresponding expressions to Eq. (29). Therefore the two point sets M′ 1, M′ 2, M′ 3, M′ 5 and m1, m2, m3, m5 are again M¨obius equivalent. As a consequence s has to be located on the cylinder of revolution Θ through M′ 1, M′ 2, M′ 3, M′ 5 with generators in direction of M4. By using again the theorem of projective geometry given at the end of the last proof we are done. □ Theorem 7. The conic q of a pentapod of Type 2 with self-motions has to be located on a cylinder of revolution, where one generator is the line g1. Proof: The proof can be done analogously to the one of Theorem 6. In order to streamline the presentation it is given in Appendix A. It should only be noted that q can be an ellipse or a circle, respectively, as q is the planar section of a cylinder of revolution. □ Theorem 8. A pentapod of Type 3 cannot possess a self-motion. Proof: In order to improve the readability of the paper the proof of the non-existence of pentapods of Type 3 with self-motions is given in Appendix B. □ 16 5.2. Second necessary condition Due to the obtained results only pentapods of Type 1,2,5 remain as candidates for self-motions. In this section we check them with respect to the second necessary condition. Theorem 9. The irreducible cubic s of a pentapod of Type 1 with self-motions has to be a straight cubic circle. Proof: Due to Theorem 6 the cubic s has three pairwise distinct points at infinity, which are denoted by M2, M3, M4. Note that M4 is real and that M2, M3 are conjugate complex; i.e. M2 = M3. The corresponding platform anchor points are denoted by m2, m3, m4 where m2 = m3 holds. Therefore we get three Darboux conditions Ωi implied by the point pairs (Mi, mi) for i = 2, 3, 4. The ideal point of the line p is again denoted by m5 and its corresponding base anchor point with M5. Therefore this point pair implies one Mannheim condition Π5. The pentapod is completed by a sphere condition Λ1 determined by the two finite points M1 and m1. The fixed frame Σ0 is chosen that M1 is the origin and that M2 and M3 are located in the xy-plane in direction (1, B2, 0) and (1, B2, 0), respectively. As M2, M3, M4 cannot be collinear, M4 is the ideal point in direction of (A4, B4, 1). Moreover we locate the origin of the moving frame Σ in m1. With respect to these coordinate systems Σ and Σ0 our conditions can be written as: Ωj : pjx0 + ajx1 + a jBjx2 + y1 + Bjy2 = 0, Ω4 : p4x0 + a4A4x1 + a4B4x2 + a4x3 + A4y1 + B4y2 + y3 = 0, Π5 : p5x0 + A5x1 + B5x2 + C5x3 + y0 = 0, (41) where (pj, 0, 0)T for j = 2, 3 are the coordinates of the intersection point of the Darboux plane and the x-axis of Σ0, (0, 0, p4)T are the coordinates of the intersection point of the Darboux plane and the z-axis of Σ0, and (p5, 0, 0)T are the coordinates of the intersection point of the Mannheim plane and the x-axis of Σ. The equation of Λ1 is given in Eq. (20) under consideration of A1 = B1 = C1 = a1 = 0. With respect to the chosen frames Σ and Σ0 the first necessary condition is fulfilled if Br = A4B4 A2 4 + 1, Bc = ± q A2 4 + B2 4 + 1 A2 4 + 1 , (42) holds with B2 = Br + iBc and Bc , 0. Then the bond β reads as follows: n0 = 0, x0 = 0, x1 = −B2, x2 = 1, x3 = A4B2 −B4, (43) y0 = B2(A5 −A4C5) −B5 + B4C5, y1 = B2a2, y2 = −a2, y3 = −a2(A4B2 −B4). (44) Now we apply the second necessary condition; i.e. the eight tangent-hyperplanes to Φ1, Φ2, Φ3, Λ1, Ω2, Ω3, Ω4, Π5 in the bond β have to have a line in common. Therefore we compute the gradients of these eight hypersurfaces with respect to the unknown n0, x0, . . . , x3, y0, . . . , y3 in the bond β. The resulting 8×9 matrix J has rank 8 (⇒β is a regular point of the 5-fold I). For the necessary condition rk(J) < 8 the determinants of all 8 × 8 submatrices of J have to vanish. The numerator of the determinant of the 8 × 8 submatrix of J, obtained by removing the column steaming from the partial derivative with respect to y2, factors into (iar −ac −ia4)(iar + ac)(A2 4 + B2 4 + 1)L1L2 with L1 = A4 q A2 4 + B2 4 + 1 ∓iB4, L2 = B5 + B5A2 4 −B4C5 −A4A5B4 ± i(A5 −A4C5) q A2 4 + B2 4 + 1, (45) and a2 = ar + iac, where ac , 0 holds. It can easily be seen that L2 can only vanish if M1, M4, M5 are collinear, which yields a contradiction. Therefore L1 = 0 has to hold, which can only be the case for A4 = B4 = 0 (⇒rk(J) = 7). This implies Br = 0 and Bc = ±i, which shows that s is a straight cubic circle. □ Theorem 10. The conic q of a pentapod of Type 2 with self-motions has to be a circle and the line g1 is orthogonal to its carrier plane (= degenerated case of a straight cubic circle). 17 Proof: The proof can exactly be done as for Theorem 9 under consideration of a4 = 0, C5 = 0 and that no two platform anchor points can coincide beside m1 and m4. It should only be noted that in this case L2 vanishes for M1 = M5, which yields a contradiction to the properties of Type 2.8 Therefore we remain again with the solution A4 = B4 = 0, which implies Br = 0 and Bc = ±i. This proves the theorem. □ Theorem 11. The cubic s∗of a pentapod of Type 5 with self-motions has to be an irreducible straight cubic circle. Proof: The proof of this theorem can be done in a similar fashion as those of Theorems 9 and 10. In order to stream- line the presentation the proof of Theorem 11 is given in Appendix C. □ Due to the Theorems 4,8,9,10,11 the condition of Duprocq [13], that the centers of the spheres have to be located on a straight cubic circle, is valid for non-planar pentapods of P. Note that Duporcq also mentioned explicitly the special cases of Type 2 and Type 5 beside the general case of Type 1. Therefore it remains to show if this so-called Duporcq condition is already sufficient for the existence of a self-motion. This is done in the next section. 5.3. Sufficiency of the Duporcq condition The sufficiency is proven separately for the Types 1,2,5. Moreover in each of the three proofs also the leg param- eters for a self-motion are given. Theorem 12. A pentapod of Type 5 fulfilling the Duporcq condition has a 1-parametric set of self-motions (over C). With respect to the coordinatization used in the proof of Theorem 11 (under consideration of A4 = B4 = 0 and B2 = i) the leg parameters are given by: w = C5 a5 , p2 = −(a3 −a5)(A5 −iB5) a5 , p3 = −(a2 −a5)(A5 + iB5) a5 , (46) and the following condition remains in R1 and R5: (a2 5 + B2 5 + C2 5)(a2 + a3 −a5) + (R2 1 −R2 5 −a2a5 −a3a5 + a2 5)a5 = 0. (47) Proof: We use the same coordinatization as given in the first two paragraphs of the proof of Theorem 11 under consideration of A4 = B4 = 0 and B2 = i. We distinguish two cases: 1. C5 −a5w , 0: Under this assumption we can solve Λ1, Ω2, Ω3, ∢4, Λ5, Φ3 for x3, y0, y1, y2, y3, n0. Plugging the obtained expressions into Φi yields the equations Φ∗ i in x0, x1, x2 for i = 1, 2. Now we compute the resultant Ξ of the numerator of Φ∗ 1 (which is quadratic in x0, x1, x2) and the numerator of Φ∗ 2 (which is quartic in x0, x1, x2) with respect to x2. Ξ factors into x4 0N[11464], where N is quartic in x0, x1. As x0 , 0 has to hold (cf. Section 4.1), N has to be fulfilled identically. Therefore we denote the coefficient of xi 0x j 1 of N by Ni j. Then N04 factors into 16E2 1E2 2 with E1 = (a3 −a5)(A5 −iB5) + a5p2, E2 = (a2 −a5)(A5 + iB5) + a5p3. (48) W.l.o.g. we can set E1 equal to zero and solve it for p2. Then the numerator of N13 factors into 32a3(a3 −a5)(A5 − iB5)(C5 −a5w)2E2 2. Therefore E2 = 0 has to hold, which can be solved for p3 w.l.o.g.. Now the numerator of N22 splits up into 64(a2 5 + B2 5)a2a3(a2 −a5)(a3 −a5)(C5 −a5w)4, (49) which cannot vanish without contradiction. 8In this case the cubic of base points splits up into three lines; one real and two conjugate complex ones, which intersect each other in M1 = M5 (complex version of Type 4). 18 2. w = C5/a5: We can solve the equations Λ1, Ω2, Ω3, ∢4, Λ5 for x3, y1, y2, y3, n0 and plug the obtained expressions into Φi, which yields quadratic equation Φ∗ i in x0, x1, x2, y0 for i = 1, 2, 3. Now then numerator of Φ∗ 1 and Φ∗ 3 do not depend on y0 in contrast to the numerator of Φ∗ 2 (coefficient of y2 0 equals 4a2 5). Therefore we compute the resultant Ξ of the numerators of Φ∗ 1 and the numerator of Φ∗ 3 with respect to x2. Ξ factors into x2 0a2 5N[208], where N is quadratic in x0, x1. We denote the coefficient of xi 0x j 1 of N again by Ni j. In the following we show that E1 = 0 and E2 = 0 has to hold: N02 factors into 4a2 5E1E2, which can be solve w.l.o.g. for p2. Then N11 splits up into −2a5E2 h a3(A2 5 + B2 5) + (a2 + a3 −a5)C2 5 + (1 −a2 −a3)a2 5 + (R2 1 −R2 5)a5 −(A5 −iB5)a5p3 i . (50) Either E2 = 0 holds and we are done or the last factor vanishes. In the latter case we can compute p3 w.l.o.g.. Then N20 factors into (C2 5 −a2 5)E2 2. For C5 = ±a5 the condition Φ1 equals x2 1 + x2 2 = 0, which cannot yield a real self-motion. Therefore E2 = 0 has to hold. Summed up we have proven that E1 = 0 and E2 = 0 have to hold. These equations can be solved for p2, p3 w.l.o.g.. Plugging the obtained expressions into N shows that only the condition given in Eq. (47) remains. This condition can always be solved for R2 5 and the self-motion is again obtained by back-substitution, which finishes the proof of the sufficiency. □ Theorem 13. A pentapod of Type 1 fulfilling the Duporcq condition has a 1-parametric set of self-motions (over C). With respect to the coordinatization used in the proof of Theorem 9 (under consideration of A4 = B4 = 0 and B2 = i) the leg parameters are given by: p2 = −A5(a2a3 −a2 4) −i(a2a3 −a2 4)B5 (a3 −a4)2 , p3 = −A5(a2a3 −a2 4) + i(a2a3 −a2 4)B5 (a2 −a4)2 , p4 = C5(a2a3 −a2 4) (a2 −a4)(a3 −a4), (51) and the following condition remains in R1 and p5: (a2 −a4)2(a3 −a4)2 h 2(a2a3 −a2 4)p5 −(a2 + a3 −2a4)R2 1 −(2a2a3 −a2a4 −a3a4)a4 i + (a2a3 −a2 4)2(a2 + a3 −2a4)(A2 5 + B2 5 + C2 5) = 0. (52) Theorem 14. A pentapod of Type 2 fulfilling the Duporcq condition has a 1-parametric set of self-motions (over C). With respect to the coordinatization used in the proof of Theorem 13 (under consideration of a4 = C5 = 0) the leg parameters are given by: p2 = −a2(A5 −iB5) a3 , p3 = −a3(A5 + iB5) a2 , p4 = 0, (53) and the following condition remains in R1 and p5: (A2 5 + B2 5)(a2 + a3) + 2a2a3p5 −R2 1(a2 + a3) = 0. (54) Proof of Theorem 13 and Theorem 14: The proofs of these theorems can be done in a fashion similar to the one of Theorem 12. For this reason and in order to streamline the presentation the corresponding proofs are given in Appendix D and Appendix E, respectively. □ Note that the proof of the sufficiency in the Theorems 12, 13 and 14 was only done over C; i.e. the self-motion has not to be real.9 Arguments of reality were only used to exclude some special cases. This can best be seen for the self-motions obtained for Type 5, as they belong to the class of Borel-Bricard motions. These are the only non-trivial motions where all points of the moving space have spherical trajectories (cf. [8, Chapter VI]; see also [29]). Note that this special case was also discussed in detail by Krames [10, Section 5]. In this case Φ∗ 1 equals x2 0(C2 5 −a2 5) + a2 5(x2 1 + x2 2) = 0. This already shows the following result: 9In contrast the conditions given in Theorem 1 and those for the cases (α, β, γ) of Section 1 are even sufficient for the existence of real self- motions. 19 Corollary 3. The 1-parametric set of self-motions given in Theorem 12 is real if |C5| < |a5| holds; complex otherwise. Until now we are not able to give a corresponding easy characterization for designs of Type 1 and 2 with real/complex self-motions. But the following two examples prove that real self-motions exist: Example 1. This example of a pentapod of Type 1 with a real self-motion is based on the formulas of Λ1, Ω2, Ω3, Ω4, Π5 given in the proof of Theorem 9. The geometry of the pentapod is determined by: a2 = B2 = i, a3 = B3 = −i, a4 = 2, A4 = B4 = 0, A5 = B5 = C5 = 1. (55) For the leg-parameters: R1 = √ 3, p2 = −3 25 −21 25i, p3 = −3 25 + 21 25i, p4 = −3 5, p5 = 46 75, (56) which are in accordance with Theorem 13, the pentapod has the following self-motion (under consideration of x0 = 1; cf. Eq. (21)): x1 = 7 4t2 −7 5t −161 300 ∓T 300, x2 = 1 4t2 −1 5t −23 300 ± 7T 300, x3 = t, (57) y1 = −1 4t2 + 1 5t + 59 300 ∓7T 300, y2 = 7 4t2 −7 5t −413 300 ∓T 300, y3 = −2t + 3 5, (58) with T = p −(75t2 −30t −41)(75t2 −90t + 31). Both branches (upper and lower one) are real for t ∈t−, t+ with t−= 1 5 −2 15 √ 33, t+ = 1 5 + 2 15 √ 33. (59) For this example we also show how to compute the finite base anchor point M (with coordinates (A, B,C)T), the finite platform anchor point m (with coordinates (a, 0, 0)T) and the leg length R of a further leg. Its corresponding sphere condition Λ has to be a linear combination of the given equations Λ1, Ω2, Ω3, Ω4, Π5; i.e. µ1Λ1 + µ2Ω2 + µ3Ω3 + µ4Ω4 + µ5Π5 −Λ = 0 (60) for any choice of n0, x0, . . . , x3, y0, . . . , y3. Therefore their nine coefficients imply nine equations in the ten unknowns µ1, . . . , µ5, R, A, B,C, a. This system has the following solution (in dependence of a): µ1 = 1, µ2 = A + Bi, µ3 = A −Bi, µ4 = 2C, µ5 = 2a, R2 = (3a2 −8a + 9)(25a4 −64a3 + 146a2 −136a + 100) 75(a2 + 1)(a −2)2 , (61) with A = a(a −1) a2 + 1 , B = a(a + 1) a2 + 1 , C = a a −2. (62) The last equation gives the bijection σ between points m of p and points M of the irreducible straight cubic circle s (cf. Eq. (16)). In Fig. 5a the trajectories of the platform anchor points m1 (a = 0), m6 (a = 1), m7 (a = 3), m8 (a = −1) and m9 (a = −2) are displayed for the upper branch of the self-motion. ⋄ Example 2. This example of a pentapod of Type 2 with a real self-motion is based on the formulas of Λ1, Ω2, Ω3, Ω4, Π5 given in the proof of Theorem 9 under consideration of a4 = 0 and C5 = 0. The geometry of the pentapod is determined by: B2 = i, B3 = −i, a2 = 1 + i, a3 = 1 −i, A4 = B4 = 0, A5 = B5 = 1. (63) For the leg-parameters: R1 = 2, p2 = −1 −i, p3 = −1 + i, p4 = 0, p5 = 1, (64) 20 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 1 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 9 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 1 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 m+ 9 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 sssssssssssssssss (a) m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 6 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 8 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m− 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 7 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m+ 6 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m− 1 = m+ 1 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 m+ 8 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M7 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M8 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M1 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M9 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 M6 qqqqqqqqqqqqqqqqq (b) Figure 5: Trajectories of the upper branch of the self-motion, where the starting pose at time t−and the end pose at time t+ are illus- trated: (a) Example 1 and (b) Example 2. Animations of these two self-motions can be downloaded from the homepage of the first author (http://www.geometrie.tuwien.ac.at/nawratil). which are in accordance with Theorem 14, the pentapod has the following self-motion (under consideration of x0 = 1; cf. Eq. (21)): x1 = −1 2t2, x2 = ±T 2 , x3 = t, y1 = 1 2t2 + 1 ∓T 2 , y2 = −1 2t2 −1 ∓T 2 , y3 = 0, (65) with T = √ −t4 −4t2 + 4. Both branches (upper and lower one) are real for t ∈t−, t+ with t−= − q 2 √ 2 −2, t+ = q 2 √ 2 −2. (66) Analogous considerations as in Example 1 show the following bijection σ between points m of p \ {m1} and points M of the circle q \ {M1}: A = a(a −2) a2 −2a + 2, B = a2 a2 −2a + 2, C = 0. (67) The point m1 is mapped to the line [M1, M9], which equals the z-axis of Σ0. In Fig. 5b the trajectories of the platform anchor points m1 = m9 (a = 0), m6 (a = 1), m7 (a = 2) and m8 (a = −1) are displayed for the upper branch of the self-motion. ⋄ Remark 6. Further examples of pentapods of Type 1 and Type 2 are implied by a remarkable motion, where all points of a hyperboloid, which carries two reguli of lines, have spherical trajectories. This well-studied motion is also known as BBM-II motion in the literature (e.g. [30]). It is known (cf. [30, page 24] and [31, page 188]) that the 21 corresponding sphere centers of lines, belonging to one regulus10, are located on irreducible straight cubic circles, which imply examples of self-motions of Type 1. Note that there also exist degenerated cases where the hyperboloid splits up into two orthogonal planes, which imply examples of self-motions of Type 2. ⋄ 6. Conclusions for practical applications We introduced a novel kinematic mapping for pentapods with linear platform (cf. Section 4.1), which can be used for an efficient solution of the direct kinematics problem. Beside this achievement we listed all pentapods with linear platform which ⋆are architecturally singular (cf. Corollary 1), ⋆possess self-motion (over C) without having an architecture singularity. The latter are either the designs (α, β, γ) given in Section 1, pentapods of Types 1,2,5 fulfilling the Duporcq condition (cf. Section 5) or the manipulator given in Theorem 1. Clearly, architecturally singular pentapods are not suited for practical application and therefore engineers should be aware of these designs. The usage of pentapods with self-motions within the design process is a double-edged sword; on the one side they should be avoided for reasons of safety11 and on the other side they have a simplified direct kinematics for the following reason: As the bonds are independent of the set of leg lengths (cf. Section 4.2), they always appear as solution of the direct kinematics problem even though the given set R1, . . . , R5 does not cause a self-motion of the manipulator. Recall that a bond of a self-motion corresponds to a complex configuration on the boundary (cf. Remark 5). Therefore four solutions (= two conjugate complex bonds of multiplicity 2; cf. Remark 5) of the direct kinematics of a pentapod with linear platform possessing a self-motion are always located on the boundary of I. This yields the following corollary: Corollary 4. A pentapod with linear platform possessing a self-motion (over C) can have a maximum of four real configurations (instead of generically 8, cf. Theorem 7), if the given set of leg lengths does not imply a self-motion. The direct kinematic problem of these pentapods reduces to the solution of a polynomial of degree 4. As this quartic equation can be solved explicitly, these pentapods seem to be of special interest for practical application. We demonstrate this result on the basis of the following example: Example 3. Continuation of Example 1: We consider the pentapod with platform anchor points m1, m6, . . . , m9 and base anchor points M1, M6, . . . , M9 of Example 1 (see Fig. 5a). But now we want to solve the direct kinematics problem of this pentapod of Type 1 for the following given set of leg lengths: R1 = 2, R6 = 1, R7 = 5, R8 = 3, R9 = 4, (68) which does not cause a self-motion. We can solve the corresponding system of equations Λ1, Λ6, . . . , Λ9 for n0, y0, y1, y2 and y3. Moreover we can set x0 = 1. Now Φ3 of Eq. (22) is only linear in x1 and x2 and we can solve it for x1. Then Φ1 and Φ2 are only quadratic in x2, and therefore the resultant of these two expressions with respect to x2 yields: 4316636297 + 69486876480x3 + 241133479200x2 3 −291209472000x3 3 + 76425120000x4 3 = 0, (69) which is only of degree 4 in x3. ⋄ But it is even possible to use this advantage of self-motions without any risk (cf. footnote 9), by designing pen- tapods with linear platform, which only have complex self-motions. In the following we list two sets of such designs: (I) Pentapods of Type 5 with |C5| ≥|a5| (cf. Corollary 3). 10The corresponding sphere centers of lines belonging to the other regulus are again located on a line (cf. [30, page 24]), which imply architec- turally singular pentapod designs. 11A self-motion is dangerous because it is uncontrollable and thus a hazard to man and machine. 22 (II) We can also solve the planar case (cf. Section 2) with the bond based approach used for the non-planar one. This study shows, that a planar pentapod of P (cf. Definition 1) has a bond if and only if the vertex V (cf. proof of Theorem 1) is an ideal point. Moreover the second necessary condition implied by the theory of bonds is only fulfilled if the affine relation (AR) holds12. Remark 7. Note that the condition (AR) equals the linear constraint given in [1], where the observation was reported that these planar pentapods with linear platform only possess a maximum of four real solutions of the direct kinematics problem (without giving an explanation for this behavior). This problem can even be solved quadratically as the solutions are symmetric with respect to the base plane (cf. [1]). Therefore this also holds for the design (γ) and for the designs (α, β) under the extra condition of a planar base. ⋄ The constraint (AR) is even sufficient for the existence of a self-motion (over C). Now we design the pentapods in a way that the distance between the parallel lines [Mi, V] and [Mj, V], which are fibers of the affinity κ, is • equal or less than the distance dist(mi, mj) between their images: This yields the pentapods characterized in Theorem 1, which all have real self-motions (cf. footnote 8). • greater than the distance dist(mi, mj) between their images: Then the line p cannot be oriented that dist(M′ i, M′ j)=dist(m′ i, m′ j) holds, which already shows that this design-set (II) is free of real self-motion. Therefore the authors recommend engineers to design pentapods with linear platform within the set (I) in the non- planar case and within the set (II) in the planar one, respectively. For reasons of completeness we also give the following corollary: Corollary 5. The following pentapods with linear platform, which do not possess a self-motion, have a maximum of six real configurations (instead of generically 8, cf. Theorem 7): 1. The irreducible cubic s of a pentapod of Type 1 is a cubic ellipse located on a cylinder of revolution. 2. The conic q of a pentapod of Type 2 belongs to a cylinder of revolution, where one generator is the line g1. 3. A pentapod of Type 5 with one of the following two properties: (a) Either the base anchor points of possible leg-replacements are located on an irreducible cubic ellipse s∗ on a cylinder of revolution. (b) or it has the following design (under consideration of footnote 1): m2 = m3, m4 = m5 and [M2, M3] is parallel to [M4, M5]. 4. Planar pentapod of P, where the associated point V is an ideal point (cf. Fig. 3c). The direct kinematic problem of the listed pentapods reduces to the solution of a polynomial of degree 6. Proof: Items 1, 2 and 3(a) are a direct consequence of the Theorems 6, 7 and 5, respectively. Item 4 follows from the above given discussion of design-set (II) and item 3(b) can be seen as its corresponding non-planar case (cf. Fig. 3c). In the latter case the cubic splits up into 3 parallel (but non-planar) lines. This case is hidden in item 2(a) of the proof of Theorem 5. Finally, it should be noted that the direct kinematics problem is only cubic for item 4 as the pentapod is planar (cf. Remark 7). □ We want to close the paper by referring to [32] where the remaining problem of characterizing designs of Type 1 and Type 2 without real self-motions is discussed, as they imply further ”save” designs with a closed form solution. 12In this case the bond (and its conjugate) are singular points of the 5-fold I (cf. Remark 5). 23 Acknowledgments The first author’s research is funded by the Austrian Science Fund (FWF): P24927-N25 - “Stewart Gough plat- forms with self-motions”. The second author’s research is supported by the Austrian Science Fund (FWF): W1214- N15/DK9 and P26607 - “Algebraic Methods in Kinematics: Motion Factorisation and Bond Theory”. References [1] Borras, J., Thomas, F., Torras, C.: A Family of Quadratically-Solvable 5-SPU Parallel Robots. Proceedings of IEEE International Con- ference on Robotics and Automation, Anchorage, Alaska, May 3-7, 2010 [2] Borras, J., Thomas, F.: Singularity-Invariant Leg Substitutions in Pentapods. Proceedings of IEEE/RSJ International Conference on Intel- ligent Robots and Systems, Taipei, Taiwan, October 18-22, 2010 [3] Weiß, G., B¨ar, G.: Singularity Investigation of a 5-leg Milling Robot. Proceedings of EuCoMeS,the first European Conference on Mech- anism Science, Obergurgl, Austria, February 21-26, 2006 [4] Pottmann, H., Wallner, J.: Computational Line Geometry. Springer (2001) [5] Gr¨unwald, A.: Die kubische Kreisbewegung eines starren K¨orpers. Zeitschrift f¨ur Mathematik und Physik 55 264–296 (1907) [6] Rulf, F.: Der kubische Kreis. Monatshefte f¨ur Mathematik und Physik 21(1) 329–335 (1919) [7] Borel, E.: M´emoire sur les d´eplacements `a trajectoires sph´eriques. M´emoire pr´esente´es par divers savants `a l’Acad´emie des Sciences de l’Institut National de France 33(1) 1–128 (1908) [8] Bricard, R.: M´emoire sur les d´eplacements `a trajectoires sph´eriques. Journal de ´Ecole Polytechnique(2) 11 1–96 (1906) [9] Husty, M.: E. Borel’s and R. Bricard’s Papers on Displacements with Spherical Paths and their Relevance to Self-Motions of Parallel Manipulators. International Symposium on History of Machines and Mechanisms (M. Ceccarelli ed.), 163–172, Kluwer (2000) [10] Krames, J.: Zur Bricardschen Bewegung, deren s¨amtliche Bahnkurven auf Kugeln liegen ( ¨Uber symmetrische Schrotungen II). Monatsheft f¨ur Mathematik und Physik 45 407–417 (1937) [11] Koenigs, G.: Le¸cons de Cin´ematique (avec notes par G. Darboux). Paris (1897) [12] Mannheim, A.: Principes et D´eveloppements de G´eom´etrie Cin´ematique. Paris (1894) [13] Duporcq, E.: Sur le d´eplacement le plus g´en´eral d’une droite dont tous les points d´ecrivent des trajectoires sph´eriques. Comptes rendus Paris 125 762–763 (1897) and Journal de math´ematiques pures et appliqu´ees (5) 4 121–136 (1898) [14] Nawratil, G., Schicho, J.: Pentapods with Mobility 2. ASME Journal of Mechanisms and Robotics 7(3) 031016 (2015) [15] Karger, A.: Architecturally singular non-planar parallel manipulators. Mechanism and Machine Theory 43(3) 335–346 (2008) [16] Nawratil, G.: Comments on “Architectural singularities of a class of pentapods”. Mechanism and Machine Theory 57 139 (2012) [17] Borras, J., Thomas, F., Torras, C.: Architectural singularities of a class of pentapods. Mechanism and Machine Theory 46(8) 1107–1120 (2011) [18] R¨oschel, O., Mick, S.: Characterisation of architecturally shaky platforms. Advances in Robot Kinematics - Analysis and Control (J. Lenarcic, M.L. Husty eds.), 465–474, Kluwer (1998) [19] Borras, J., Thomas, F.: Kinematics of Line-Plane Subassemblies in Stewart Platforms. Proceedings of IEEE International Conference on Robotics and Automation, Kobe, Japan, May 12-17, 2009, pages 4094–4099 [20] Nawratil, G.: Self-motions of planar projective Stewart Gough platforms. Latest Advances in Robot Kinematics (J. Lenarcic, M. Husty eds.), 27–34, Springer (2012) [21] Nawratil, G.: Introducing the theory of bonds for Stewart Gough platforms with self-motions. ASME Journal of Mechanisms and Robotics 6(1) 011004 (2014) [22] Nawratil, G.: Correcting Duporcq’s theorem. Mechanism and Machine Theory 73 282-295 (2014) [23] Sch¨onflies, A.: Geometrie der Bewegung in synthetischer Darstellung. Teubner, Leipzig (1886) [24] Husty, M., Pfurner, M., Schr¨ocker, H.-P., Brunnthaler, K.: Algebraic methods in mechanism analysis and synthesis. Robotica 25 (6) 661–675 (2007) [25] Husty, M., Schr¨ocker, H.-P.: Kinematics and algebraic geometry. 21st Century Kinematics - The 2012 NSF Workshop (J.M. McCarthy ed.), 85–123, Springer (2012) [26] Husty, M.L.: An algorithm for solving the direct kinematics of general Stewart-Gough platforms. Mechanism and Machine Theory 31(4) 365–380 (1996) [27] Zhang, C.-de, Song, S.-M.: Forward kinematics of a class of parallel (Stewart) platforms with closed-form solution. Proceedings of IEEE International Conference on Robotics and Automation, Sacramento, California, April 9-11, 1991, pages 2676–2681 [28] Gallet, M., Nawratil, G., Schicho, J.: Bond Theory for Pentapods and Hexapods. Journal of Geometry 106(2) 211–228 (2015) [29] Lee, C.-C., Herv´e, J.M.: Bricard one-DoF motion and its mechanical generation. Mechanism and Machine Theory 77 35–49 (2014) [30] Hartmann, D.: Singular Stewart-Gough Platforms. Master Thesis, Department of Mechanical Engineering, McGill University, Montreal, Canada (1995) [31] Krames, J.: Die Borel-Bricard-Bewegung mit punktweise gekoppelten orthogonalen Hyperboloiden ( ¨Uber symmetrische Schrotungen VI). Monatsheft f¨ur Mathematik und Physik 46 172–195 (1937) [32] Nawratil, G.: On the line-symmetry of self-motions of linear pentapods. [arXiv:1510.03567] 24 Appendix A: Proof of Theorem 7 We can start with the same set of equations as in the proof of Theorem 6, but we have the extra conditions m1 = m4 and that M1, M2, M3, M5 are located in a plane, which does not contain M4. Due to the properties of Type 2 not all base points are coplanar (⇒B2C3C5 , 0) and no two platform anchor points can coincide beside m1 and m4 (⇒ a1 = a4 = 0). Therefore we can express the coplanarity of M1, M2, M3, M5 as A3 = A2B3C5 −A2B5C3 + B2C3A5 B2C5 . (70) We set x0 = 0 and start the computation of the bonds: We can solve the equations Λ1, Λ2, Λ3, Ω4, Π5 for y0, y1, y2, y3, n0 w.l.o.g.. Now the numerator of Γ4 factors into x1F with F of Eq. (36). Therefore we have to dis- tinguish two cases: 1. x1 , 0: In this case F = 0 has to hold, which can be solve w.l.o.g. for x3. Then the numerator of Γ6 factors into: x2a3(x2B2 + x1A2)(B3C5 −B5C3 + B2C3 −B2C5). (71) The last factor cannot vanish, as otherwise the base anchor points M2, M3, M5 are collinear, a contradiction. There- fore we remain with two cases: a. x2 = −x1A2/B2. Now the numerator of Γ1 factors into: x2 1 h A2 2(B2 5 + C2 5) −2A2A5B2B5 + B2 2(A2 5 + C2 5) i . (72) The discriminant with respect to A2 equals −C2 5(A2 5 + B2 5 + C2 5) and therefore we get a contradiction. b. x2 = 0: In this case the numerator of Γ1 factors into x2 1[C2 5 + (A2 −A5)2] which can also not vanish without contradiction. 2. x1 = 0: This case is exactly the same as the one discussed in item 2 of the proof of Theorem 6, which already yields the result. □ Appendix B: Proof of Theorem 8 We can make leg-replacements such that m1 = m2 = P1, m3 = m4 = P2, M1, M2 ∈L1 and M3, M4 ∈L2. Moreover we can choose M2 and M3 as ideal points of L1 and L2, respectively. Therefore the point pairs (Mi, mi) determine sphere conditions Λi for i = 1, 4 and Darboux condition Ωi for i = 2, 3. Moreover we can assume that M1 is located in the Darboux plane of (M2, m2) and that M4 is located in the Darboux plane of (M3, m3). Finally we can assume that m5 is the ideal point of the line p. Therefore the point pairs (M5, m5) determines a Mannheim condition Π5. W.l.o.g. we choose the fixed frame Σ0 that M1 equals its origin and that M2 and M3 are located in the xy-plane symmetric with respect to the x-axis. Therefore the directions of M2 and M3 are given by (1, B2, 0) and (1, −B2, 0), respectively. Moreover we can define the moving frame Σ in a way that m1 is its origin. With respect to these coordinate systems our conditions can be written as: Ω2 : y1 + B2y2 = 0, Ω3 : (A4 −B2B4)x0 + a3x1 −a3B2x2 + y1 −B2y2 = 0, Π5 : p5x0 + A5x1 + B5x2 + C5x3 + y0 = 0, (73) where (p5, 0, 0)T are the coordinates of the intersection point of the Mannheim plane and the x-axis of Σ. The equations of Λi for i = 1, 4 are given in Eq. (20) under consideration of A1 = B1 = C1 = a1 = 0 and a4 = a3. Now we set x0 = 0 and prove that no bonds can exist. W.l.o.g. we can solve Λ1, Ω2, Ω3, Π5 for n0, y0, y1, y2. Then the numerator of Γ4 can only vanish in the following two cases: 25 1. x1 = B2x2: The numerator of Γ2 implies y3 = 0. Now the numerator of Λ4 equals: a4 [x2(B2A4 −B2A5 + B4 −B5) + (C4 −C5)x3] . (74) We distinguish two cases: (a) C4 , C5: Under this assumption we can solve the last factor for x3. Then x2 2 factors out from the numerator of Γ1 and we remain with only one condition, which is quadratic with respect to A5. The corresponding discriminant equals −(C4 −C5)2(B2 2 + 1) and therefore no solution exists. (b) C4 = C5: In this case Λ4 can only vanish for x2 = 0 as B2A4 −B2A5 + B4 −B5 = 0 implies the collinearity of M3, M4, M5, a contradiction. Then Γ1 cannot vanish without contradiction. 2. x1 = −B2x2: Now the numerator of Γ6 equals −x2(x3a4 + y3). Therefore we have to distinguish two cases: (a) x2 = 0: Now Γ1 and Γ2 imply x3 = y3 = 0, a contradiction. (b) y3 = −x3a4: Then Λ4 factors into a4 [x2(B2A5 −B5) −C5x3] . (75) We distinguish two cases: i. C5 , 0: Under this assumption we can solve the last factor for x3. Then x2 2 factors out from the numerator of Γ1 and we remain with only one condition, which is quadratic with respect to A5. The corresponding discriminant equals −C2 5(B2 2 + 1) and therefore no solution exists. ii. C5 = 0: In this case Λ4 can only vanish for x2 = 0 as B2A5 −B5 = 0 implies the collinearity of M1, M2, M5, a contradiction. Then Γ1 cannot vanish without contradiction. □ Appendix C: Proof of Theorem 11 Due to Theorem 5 the irreducible cubic s∗has three pairwise distinct points at infinity, which are denoted by M2, M3, M4. Note that M4 is real and that M2, M3 are conjugate complex; i.e. M2 = M3. The corresponding platform anchor points are denoted by m2, m3, m4 where m2 = m3 holds. Therefore we get two Darboux conditions Ωi implied by the point pairs (Mi, mi) for i = 2, 3. Moreover the point pair (M4, m4) implies the angle condition ∢4. The pentapod is completed by two sphere conditions Λj, which are determined by the two finite points Mj and mj for j = 1, 5. We choose the fixed frame Σ0 that M1 is the origin and that M2 and M3 are located in the xy-plane in direction (1, B2, 0) and (1, B2, 0), respectively. As M2, M3, M4 cannot be collinear, M4 is the ideal point in direction of (A4, B4, 1). Moreover we can define the moving frame Σ in a way that m1 is its origin. With respect to these coordinate systems our conditions can be written as: Ωi : pix0 + aix1 + aiBix2 + y1 + Biy2 = 0, ∢4 : wx0 + A4x1 + B4x2 + x3 = 0 (76) where (pi, 0, 0)T for i = 2, 3 are the coordinates of the intersection point of the Darboux plane and the x-axis of Σ0 and w denotes arccos (ϕ). The equations of Λj is given in Eq. (20) under consideration of A1 = B1 = C1 = a1 = 0 for j = 1, 5. In the following we substitute B2 = Br + iBc and a2 = ar + iac with Br, Bc, ar, ac ∈R and Bcac , 0. Then we set x0 = 0 and start the computation of bonds: We can solve the equations Λ1, Ω2, Ω3, ∢4, Λ5 for x3, y1, y2, y3, n0 w.l.o.g.. Now the numerator of Γ4 can only vanish without contradiction for x1 = (−Br ± iBc)x2. We only discuss the upper sign as the conjugate solution can be done analogously. Then the numerator of Γ6 can be solved for y0 w.l.o.g.. From the numerator of Γ1 we can factor out x2 2 and we remain with only one condition. From its imaginary part and real part we can compute Br and Bc given in Eq. (42). Then we compute again the 8 × 9 matrix J with respect to the obtained bond. For the necessary condition rk(J) < 8 the determinants of all 8 × 8 submatrices of J have to vanish. The numerator of the determinant of the 8 × 8 submatrix of J, obtained by removing the column steaming from the partial derivative with respect to x2, factors into (iar +ac −ia5)(iar +ac)(A2 4 + B2 4 +1)L1L2 with L1 and L2 of Eq. (45). Analogous arguments as in the proof of Theorem 9 shows that s∗has to be a straight cubic circle. □ 26 Appendix D: Proof of Theorem 13 We use the same coordinatization as given in the first three paragraphs of the proof of Theorem 9 under consider- ation of A4 = B4 = 0 and B2 = i. W.l.o.g. we can solve the equations Λ1, Ω2, Ω3, Ω4, Π5 for y0, y1, y2, y3, n0. Plugging the obtained expressions into Φi yields quadratic equation Φ∗ i in x0, . . . , x3 for i = 1, 2, 3. Therefore Φ∗ i corresponds with a quadric in the homogeneous 3-space (spanned by x0, . . . , x3 ). In the following we want to determine the parameters R1, p2, p3, p4, p5 in a way that these three quadrics have a curve in common. This can be done as follows: We compute the resultant Ξk of Φ∗ i and Φ∗ j with respect to x1 for pairwise distinct i, j, k ∈{1, 2, 3}. Ξ1, Ξ2 and Ξ3 are homogeneous quartic expressions in x0, x2, x3, but they are only quadratic with respect to x2. Therefore we eliminate this unknown by computing the resultant Υk of Ξi and Ξj for pairwise distinct i, j, k ∈{1, 2, 3}. Then the greatest common divisor N of Υ1, Υ2 and Υ3 has to vanish. It turns out that N has 411 terms and that it is homogeneous of degree 4 in x0, x3. For a self-motion of the line p the expression N has to be fulfilled independently of x0, x3. Therefore we denote the coefficient of xi 0x j 3 of N by Nij. In the following we show that the three conditions E3 = 0, E4 = 0, E5 = 0 have to be fulfilled with: E3 = A5(a2a3 −a2 4) + p2(a3 −a4)2 −i(a2a3 −a2 4)B5, E4 = A5(a2a3 −a2 4) + p3(a2 −a4)2 + i(a2a3 −a2 4)B5, E5 = C5(a2a3 −a2 4) −p4(a2 −a4)(a3 −a4). (77) This can be seen as follows: N04 = 0 splits up into E3E4. W.l.o.g. we can set E3 equal to zero and solve it for p2. Then the numerator of N13 factors into −2a3(A5 −iB5)E4E5. Therefore we have to distinguish two cases: 1. E4 = 0: W.l.o.g. we can solve this equation for p3. Then the numerator of N22 splits up into 4a2a3(A2 5 + B2 5)E2 5. Therefore E5 = 0 has to hold and we are done. 2. E5 = 0: W.l.o.g. we can solve this equation for p4. Then the numerator of N22 and N31 splits up into: a3(A5 −iB5)E4F22[83], a3a4C5(A5 −iB5)E4F31[73]. (78) Either E4 = 0 holds and we are done or F22 = 0 and F31 = 0 have to hold. In the latter case we compute F22 + F31, which factors into a3(a2 −a4)2(A5 −iB5)E4. Therefore again E4 = 0 has to be fulfilled. Summed up we have proven that E3 = 0, E4 = 0, E5 = 0 have to hold. These equations can be solved for p2, p3, p4 w.l.o.g.. Plugging the obtained expressions into N shows that only the condition given in Eq. (52) remains. We distinguish two cases: • a2a3 −a2 4 , 0: In this case Eq. (52) can always be solved for p5. Then the self-motion can be computed by back-substitution; i.e. we compute the common factor of Ξ1, Ξ2, Ξ3 and solve it for x2 (which only appears quadratic) and finally the common factor of Φ∗ 1, Φ∗ 2 and Φ∗ 3 (which is linear in x1) gives the self-motion. • For the special case a2a3 −a2 4 = 0 we get R2 1 = a2 4 from Eq. (52). Moreover if we set a2 = ar +iac with ar, ac ∈R and ac , 0 the condition a2a3 −a2 4 = 0 is equivalent with a2 r + a2 c −a2 4 = 0, which can be solved for a4 w.l.o.g.. Now it can easily be seen that Φ∗ 1(a2 r + a2 c) = Φ∗ 2 holds. Therefore we only remain with the following condition beside Φ∗ 1 (which equals Φ1 given in Eq. (22)): Φ∗ 3 : x2 0p5 −(x2 1 + x2 2)ar −x2 3 q a2r + a2c + x0(x1A5 + x2B5 + x3C5) = 0. (79) This finishes the proof of the sufficiency. □ 27 Appendix E: Proof of Theorem 14 We start with the same set of equations as in the proof of Theorem 13 under consideration of a4 = 0, C5 = 0 and that no two platform anchor points can coincide beside m1 and m4. Moreover we can compute analogously the corresponding expression N, which has in this case only 204 terms. We denote the coefficient of xi 0x j 3 of N again by Ni j. In the following we show that the three conditions E6 = 0, E7 = 0 and p4 = 0 have to be fulfilled with: E6 = a2(A5 −iB5) + a3p2, E7 = a3(A5 + iB5) + a2p3. (80) This can be seen as follows: N04 = 0 splits up into a2a3E6E7. W.l.o.g. we can set E6 equal to zero and solve it for p2. Then the numerator of N13 factors into 2a2 2(A5 −iB5)p4E7. Therefore we have to distinguish two cases: 1. E7 = 0: W.l.o.g. we can solve this equation for p3. Then the numerator of N22 splits up into 4a2a3(A2 5 + B2 5)p2 4. Therefore p4 = 0 has to hold and we are done. 2. p4 = 0: Now the numerator of N22 splits up into: a2(A5 −iB5)E7 h a2p3(A5 −iB5) + R2 1(a2 + a3) −a2(A2 5 + B2 5 + 2a3p5) i . (81) If E7 = 0 holds we are done. Therefore we express p3 from the last factor, which can be done w.l.o.g.. Then N31 is fulfilled identically and we only remain with N40. Its numerator factors into: a2R2 1 h 2a2a3p5 + (a2 + a3)(A2 5 + B2 5 −R2 1) i . (82) Now we have to distinguish two cases: (a) We can solve the last factor for p5 w.l.o.g.. Then it can easily be checked that E7 is fulfilled and we are done. (b) R1 = 0: This condition implies a spherical self-motion of the line p with center M1 = m1. Now a one- parametric set of legs with base anchor points on the circle q (the corresponding platform anchor points on p are given by σ−1) can be attached without restricting the spherical self-motion. It can easily be seen that any two legs of this set already fix the pose of p; hence no real self-motion of p exists. Summed up we have proven that E6 = 0, E7 = 0 and p4 = 0 have to hold. E6 = 0, E7 = 0 can be solved for p2, p3 w.l.o.g.. Plugging the obtained expressions into N shows that only the condition given in Eq. (54) remains. This condition can always be solved for p5 and the self-motion is again obtained by back-substitution, which finishes the proof of the sufficiency. □ 28