Method and device for controlling movements in the case of manipulators

Abstract

In a method for controlling the movement of a manipulator, such as an industrial robot, associated with an interpretation of a given point sequence of poses (positions and orientations) by splines, the motion components are separately parametrized. Thus, marked, subsequent changes to the orientation of robot axes have no undesired effects on the Cartesian movement path of the robot. The invention also proposes suitable algorithms for orientation control by means of quaternions or Euler angles.

Description

FIELD OF THE INVENTION

[0001] The invention relates to a method for controlling movements of a manipulator, such as an industrial robot, comprising an interpolation of a given point sequence of poses (positions and orientations) by splines and a device, more particularly suitable for performing the method according to the invention.

BACKGROUND OF THE INVENTION

[0002] Movements of controlled manipulators, e.g. industrial robots, are generally determined by a preestablished sequence of points in a multidimensional space. In the case of a multiaxial industrial robot each of these points incorporates the Cartesian coordinates of positions and coordinates with respect to the orientation in three-dimensional Euclidean space and optionally coordinates of auxiliary axes.

[0003] As a rule a robot movement is not rigidly predetermined, but instead merely has a number of fixed points, which are based on the operations to be performed, e.g. the taking up and setting down of workpieces, passing round obstacles, etc. The fixed points are taught to the robot control used for movement control purposes in a manual manner by starting and storing the corresponding poses and are then available as a point sequence with respect to which the robot movement to be executed must be oriented, i.e. all the fixed points clearly define the resulting path curve.

[0004] In order to obtain a handy, functional description in place of a point sequence and permit the continuity of the robot movement in all components of motion, the point sequence is generally interpolated. A functional description of the point sequence obtained by interpolation serves a number of purposes. Firstly the data quantity is compressed and secondly it is possible to obtain informations with respect to values not contained in the point sequence, e.g. by extrapolation. Finally, a higher evaluation rate occurs, because the evaluation of a function can often be implemented faster than a reference in a table.

[0005] In the industrial practice of movement control for interpolating the aforementioned point sequences use is frequently made of spline functions, which satisfy tailor-made demands, particularly with respect to the smoothness of the curve obtained, the continuity and differentiatability of the first derivative thereof and the continuity of the second derivative thereof. As the first and second derivatives of a path curve can be identified with the speed or acceleration of an object moved along the curve, said characteristics are of great importance for an efficient movement control of industrial robots.

[0006] A known method for the movement control of industrial robots using spline interpolations is described in an article by Horsch and Jüttler (Computer-Aided Design, vol. 30, pp 217-224, 1998). The contour of a robot movement is represented sectionwise by means of a suitable degree or order polynomial.

[0007] In the known method for the movement control of industrial robots by spline interpolation, it has proved disadvantageous that in sections of the point sequence to be interpolated in which changes to the orientation, the Cartesian position and possibly the position of the additional axes occur in widely differing form (e.g. minimum Cartesian changes with large orientation changes), undesired, uneconomic movement paths can occur. In particular, in such cases movement loops can occur, i.e. one or more components of the robot movement locally lead to an inefficient moving backwards and forwards, because the movement guidance in this specific section of the point sequence is dominated by significant changes in other components of motion. It is also considered disadvantage with per se fixed movement paths a subsequent pronounced change to individual components of motion, e.g. the subsequent programming of a strong reorientation, can have an effect on the behaviour of the other components of motion, so that e.g. as a consequence of the reorientation the Cartesian contour of the path also changes, which is unacceptable for the user.

[0008] For the orientation guidance of Cartesian movements, which are e.g. in the form of splines, frequent use is made of quaternions. Quaternions are generalized complex numbers in the form of a number quadruple, which can be represented as a real-value scalar part and complex-value three-vector. Quaternions are closely linked with the matrix representation of rotations and are therefore preferably used for the orientation guidance in computer animations and for robot control purposes. A point sequence of robot orientations is consequently represented by a four-dimensional spline in the quaternion space.

[0009] In order to obtain in this connection an optimum favourable, uniform parametrization of the orientation movement, the spline in the quaternion space should as far as possible be on the surface of the unit sphere. It is only in this way possible to achieve a uniform parametrization of the projection of the quaternion spline on the unit sphere, because on converting quaternions into rotation matrixes each quaternion qi must be standardized. Through the reduction of the four-dimensional quaternion spline to three degrees of freedom during projection on the unit sphere (standardization of the quaternion spline or the individual quaternions), the risk arises of a distorted parametrization of the standardized spline, which can in turn lead to an undesired movement behaviour of a robot controlled in this way.

[0010] In the known spline interpolation method there is a parametrization of the movement or spline by means of a common parameter, often called the motion parameter t and which is related to the time coordinate τ (cf. Horsch and Jüttler, top left on p 221). For a given point sequence Pi, i=1, . . . , n, the length of the parametrization interval [ti,ti+1] is generally considered to be proportional to the spacing of the associated points Pi,Pi+1. Into said spacing is introduced both the Cartesian distance and changes to the orientation and optionally auxiliary axes. However, it has been found that a parametrization by means of only a single parameter gives to the aforementioned disadvantages.

[0011] The problem of the invention is to further develop a method and device of the aforementioned type in such a way that there are no undesired, uneconomic path movements, such as loops or the like. In addition, movement or motion control is to take place in such a way that reorientation from point to point is at a minimum.

SUMMARY OF THE INVENTION

[0012] In the case of the method of the aforementioned type., the set problems are solved by the invention in that components of the motion are separately parametrized. In a device of the aforementioned type, the set problems are solved in that it has an interpolation device, which is constructed for the separate parametrization of components of motion during interpolation.

[0013] According to the invention, through the separate parametrization of all the components of Pi, which must not influence one another, e.g. the Cartesian position and orientation, the aforementioned disadvantages are avoided.

[0014] According to a further development of the invention, the Cartesian positions of the manipulator are parametrized or parametrizable separately from the associated orientations.

[0015] If the manipulator to be controlled has auxiliary axes, according to a further development of the invention, movements of auxiliary axes are parametrized or parametrizable separately from the positions and/or orientations of the manipulator.

[0016] In order to utilize the known, suitable characteristics of quaternions in the description of rotary movements in space, according to a preferred development of the invention, an orientation guidance takes place by means of quaternions qi, i=1, . . . , n and the device according to the invention is set up for such an orientation guidance.

[0017] Alternatively, according to a further development of the invention, an orientation guidance takes place by means of successive rotations with Euler angles Ai, Bi, Ci, i=1, . . . , n, and the device according to the invention is set up for such an orientation guidance and in each case directly succeeding rotations take place about axes perpendicular to one another.

[0018] Preferably, in this connection, use is made of extended or expanded Euler angles with Ai, Bi, Ci∈R (R: quantity of real numbers).

[0019] In this way, according to the invention, a uniform parametrization of the orientation is always ensured and for the representation thereof only three degrees of freedom are required. In this connection, the standard Euler angles are inventively expanded to the total R3, which ensures minimum reorientations. It has surprisingly been found that by removing the limitation of the definition range of Euler angles, much more economic movement paths can be determined.

BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Further advantages and features of the invention can be gathered from the claims and the subsequent description of specific embodiments relative to the attached drawings, wherein show:

[0021] FIG. 1 a a path curve of a robot movement in two dimensions according to the prior art with undesired loop formation.

[0022] FIG. 1 b A path curve according to the invention of a robot movement in two dimensions without loop formation.

[0023] FIG. 2 a A rotation in the description with Euler angles.

[0024] FIG. 2 b The rotation according to FIG. 2a in a different description with Euler angles.

[0025] FIG. 3 A rotation movement in the representation with standard Euler angles.

[0026] FIG. 4 A representation of the rotation of FIG. 3 with extended or expanded Euler angles.

[0027] FIG. 5 A flow chart of a method according to the invention when using the quaternion algorithm.

[0028] FIG. 6 A flow chart of an inventive method when using the Euler angle algorithm.

[0029] FIG. 7 A block diagram of the device according to the invention.

DETAILED DESCRIPTION OF THE DRAWINGS

[0030] In a simplified, two-dimensional representation, FIG. 1a shows a Cartesian motion contour B for the motion of an industrial robot. It is possible to see two Cartesian coordinates x(t), y(t), which are dependent on a common parameter t, which is generally related to a time coordinate. At S the path contour B has a loop, i.e. over the associated interval of parameter t, the robot performs a generally undesired, because uneconomic forward and backward movement in the x and/or y direction and which should be avoided in practice. Curve B interpolates a predetermined sequence of points Pi, which give the course of the robot movement in a multidimensional space. The points Pi can be taught the robot control responsible for the movement control or this is obtained online during a manipulation process, e.g. from sensor informations or the like. For interpolating Pi during robot movements use is made of standard splines (Horsch and Jüttler, loc. cit.). Hereinafter

P i =(Xi,Yi,Zi,Ai,Bi,Ci,El, . . . ,Em)∈R6+m, i=1, . . . ,n

[0031] are the robot positions to be interpolated (for a 6 axis robot with m auxiliary axes E1 to Em). Thus, what is sought is a parametrization of a spline Sp of the desired contour, i.e.

Sp (ti)=Pi for all i=1, . . . ,n and ti<t2<t3< . . . <tn,

[0032] which leads to an optimum smooth curve. In order to obtain a favourable contour, the length of the parametrization intervals [ti,ti+1] is generally chosen proportional to the spacing of the associated points Pi,Pi+1. Into said spacing must be introduced both the Cartesian distance and the orientation change and possibly auxiliary axis changes. This procedure is intended to avoid undesired loops.

[0033] A spline is generally a piecemeal polynomial interpolation with selectable characteristics of the polynomials at the junction points (e.g. continuity and differentiatability of the first derivative and/or continuous second derivative in the case of cubic spline interpolation with third degree or cubic functions).

[0034] A parametrization with only one parameter t can, as shown in FIG. 1a, in sections [Pi,Pi+1] where modifications to the orientation, the Cartesian position and possibly the auxiliary axes differ widely (e.g. minimum Cartesian modifications with simultaneously a large orientation change), lead to a parameter interval [ti,ti+1], whose length is not simultaneously favourable for orientation guidance and the Cartesian position. In such cases loops can still occur (FIG. 1a).

[0035] Subsequently programmed, marked reorientations can also change, e.g. lengthen the common parameter interval [ti,ti+1], so that the Cartesian contour changes, which is not acceptable in practice.

[0036] When interpolating a given sequence of points Pi through a single, continuous curve, in addition to the “taught” Cartesian positions (X, Y, Z), also the orientations (e.g. represented by means of Euler angles (A, B, C) and corresponding auxiliary axis positions (E1, E2, . . . )) are interpolated and simultaneously the reorientation from point to point is to be minimized. In addition, a pure change to the orientation must not modify the Cartesian curve or auxiliary axis contour and vice versa. Thus, the invention proposes the separate parametrization of all components of Pi, which cannot influence one another. For example, it is possible to separately parametrize the Cartesian position, orientation (independently of a description with Euler angles or quaternions) and the different auxiliary axes, e.g.:

Kar(ti):=(Xi,Yi,Zi), i=1, . . . n; ti ∈R, ti<t2< . . . <tn

Orio(oi):=(Ai,Bi,Ci), i=1, . . . n; oi ∈R, o1<o2< . . . <on

Ek (ei):=Eki, i=1, . . . , n, k=1, . . . ,m; eki ∈R, eki<ek2< . . . <ekn.

[0037] Through a monotonically increasing intermediate parametrization

o (t) with o(ti)=oi for all i=1, . . . , n

ek (t) with ek(ti)=eki for all i=1, . . . , n, k=1, . . . , m

[0038] the desired interpolation characteristic is guaranteed and a contour shown in FIG. 1b is obtained satisfying the desired requirements:

Sp (t)=(Kar(t), Orio(o(t)), E=(e=(t)), . . . , Em(em(t))).

[0039] In order to maintain monotony of the intermediate parametrization for a given order of differentiatability, it is e.g. possible to make use of interpolating splines, which contain an additional parameter by means of which the spline can be forced randomly close to the associated polygonal course (cf. e.g. Helmuth Späth, Spline Algorithms, 4th edition, R. Oldenbourg Verlag 1986, Generalized Cubic Spline Functions).

[0040] A pure orientation change in one point consequently only changes the parametrization or intermediate parametrization of the orientation, but not that of the Cartesian path or the auxiliary axes.

[0041] Orientation guidance can take place with quaternions or Euler angles.

[0042] 1. Orientation Guidance with Quaternions

[0043] Each reorientation in space can be represented by a rotation about a suitable vector v∈R3 (with |v|=1) and angle φ. Between this pair (vector, angle) and the associated quaternion q, the following link exists: 1$\begin{array}{c}q=\left(w,x,y,z\right)\\ =(\cos \left(1/2\phi \right),\sin \left(=/2\phi \right)·v\\ =\left(\cos \left(-1/2\phi \right),\sin \left(-1/2\phi \right)·\left(-v\right)\right)\end{array}$

[0044] Thus, a rotation by −v with −φ not only leads to the same orientation, but also to the same quaternion.

[0045] However, a rotation by v in the opposite direction, i.e. with −(2π−φ), naturally also leads to the same orientation, but −q is obtained as the associated quaternion (the sign contains the information concerning the rotation direction). Between the pair (rotation vector, angle) and the associated quaternions, the following association consequently exists:

q (V,φ), (−v,−φ)

−q(v, −(2π−φ)), (−v, 2π−φ).

[0046] For the unit quaternions q=(±1, 0, 0, 0) φ=0 or 2π, and due to sin(½φ)=0 the vector v can be chosen at random. For all remaining quaternions on the unit circle it is always possible to determine two pairs (v,φ), (−v,−φ) with φ≠0 or φ≠2π.

[0047] On converting into an orientation matrix the quaternion is standardized, i.e. all positive multiples of a quaternion differing from zero lead to the same orientation matrix and rotation direction.

[0048] In order to discover an absolutely uniformly parametrized orientation control by means of a quaternion spline, which is always located on the surface of the unit sphere in quaternion space and where the dimension of the orientation spline is reduced by one dimension (from four to three), the invention proposes that for the determination of the spline successively the following steps are performed (hereinafter N is the quantity of the natural numbers and Z the quantity of the whole numbers):

[0049] Representation of each quaternion on the four-dimensional unit sphere according to

q (α,β,φ)=(cos(½φ), sin(½φ)·v(α,β))

[0050] with

(v(α,β)=(cos(α) cos(β), sin(α) cos(β), sin(β)),

[0051] in which α, β, (and r=1) give the spherical coordinates of the rotating vectors v=v(α,β)

[0052] determination of motion-optimum pairs (vi,φi) by

[0053] a) adapting the sign of qi, so that

q iT q i+1 ≧0 for all i=1, . . . , n−1;

[0054] b) for all i=1, . . . , n determination of a possible pair (vi, φi) with

q i =(cos(½φi), sin(½φi)vi);

[0055] c) choice of a random starting pair (ViStart, φiStart), with iStart∈I:={i=1, . . . ,n | qi≠(±1, 0, 0, 0)}, e.g. iStart=min{i∈I};

[0056] on choosing iStart∉I, the completely freely selectable rotary vector viStart would be unnecessarily fixed, which can lead to a suboptimum orientation, because the starting pair would itself no longer be adapted;

[0057] d) for all j∈{1, . . . , n} with the exception of j=iStart determination of in each case precisely one pair (vj,φj starting from a reference pair (vi,φi) in accordance with steps e) to g), in which i∈{1, . . . , n) is a neighbouring index, i.e. i=j+l or i=j−1 and also either i=iStart or the reference pair has already been adapted according to steps e) to g);

[0058] e) determination of k*, h*∈Z, so that 2$\begin{array}{c}\&LeftBracketingBar;{\phi }_{j,1}(k\mathrm{*)}-{\phi }_i\&RightBracketingBar;=\min \left\{\&LeftBracketingBar;{\phi }_{j,1}\left(k\right)-{\phi }_i\&RightBracketingBar;,k\in Z\right\}\\ =\min \left\{\&LeftBracketingBar;{\phi }_j+4k\pi -{\phi }_i\&RightBracketingBar;,k\in Z\right\},\\ \&LeftBracketingBar;{\phi }_{j,2}(h\mathrm{*)}-{\phi }_i\&RightBracketingBar;=\min \left\{\&LeftBracketingBar;{\phi }_{j,2}\left(h\right)-{\phi }_i\&RightBracketingBar;,h\in Z\right\}\\ =\min \left\{\&LeftBracketingBar;-{\phi }_j+4h\pi -{\phi }_i\&RightBracketingBar;,h\in Z\right\};\end{array}$

[0059] f) subsequently for j∈I determination of

d j,1 =|arccos(viTj)|+|φj,1(k*)−φi|,

d j,2 =|arccos(viT(−vj))|+|(φj,2(h*)−φi−

or for j∈I

d j,1 =|φj,1(k*)−φi|,

d j,2 =|φj,2(h*)−φi|;

[0060] g) subsequently in each case choice of (vj, φj), where for 3$j\notin I\left(v_j,{\phi }_j\right)=\begin{array}{c}\{\left(v_j,{\phi }_{j,1}\left(k^{*}\right)\right),d_{j,1}\le d_{j,2}\\ \{\\ \{\left(-v_j,{\phi }_{j,2}\left(h^{*}\right)\right),d_{j,1}>d_{j,2}\end{array}$$\mathrm{or}\mathrm{for}j\notin I\left(v_j,{\phi }_j\right)=\begin{array}{c}\{\left(v_i,{\phi }_{j,1}\left(k^{*}\right)\right),\mathrm{if}{d}_{j,1}\le d_{j,2}\\ \{\\ \{\left(v_i,{\phi }_{j,2}\left(h^{*}\right)\right),\mathrm{if}{d}_{j,1}>d_{j,2};\end{array}$

[0061] vj can be in principle randomly selected for j∉I. The specific choice (vj=vi) is only provisional (until all (vj, φj) are adapted) and guarantees that in step d) there is always a complete reference pair and the freely selectable rotary vectors are ignored (i.e. skipped), so that finally there is always a minimization of the spacing between neighbouring, not freely selectable rotary vectors.

[0062] When all the pairs (vi, φi) are adapted according to steps e) to g), the freely selectable vi (with i∈I) are no longer required and can be deleted or cancelled;

[0063] determination of pairs (αi, βi) from vi in spherical coordinate representation by:

[0064] h) determining standard spherical coordinates (αi, βi) of vi for all i∈I:

v i =(cos(αi) cos(βi), sin(αi) cos(βi), sin(βi));

[0065] i) definitions of

J:={j∈I with vj≠(0, 0, ±1)},

L:={j∈I with vj=(0, 0, ±1)}

[0066] and therefore definition of expanded spherical coordinates from R×R

[0067] for i∈J:

(αi,1(h), βi,1(k))=(αi+2hπ, βi+2kπ), h, k∈Z;

(αi,2(1), βi,2(m))=(αi+π+2lπ, π−βi+2mπ), l, m∈Z;

[0068] for i∈L:

(αi(x), βi(k))=(x, βi+2kπ), x∈R, k∈Z

[0069] (as βi=±π/2 for i∈L, supplies (αi(x)+π, π−βi(k)) no further solutions);

[0070] j) fixing a random starting pair (αiStart, βiStart) with jStart∈J, e.g. jStart=min {j∈J} (jStart∈L would be unfavourable, because the freely selectable parameter αj would then be fixed beforehand, which would lead to a suboptimum solution);

[0071] k) for all j∈I with the exception of j=jStart in each case determination of expanded polar coordinates in accordance with steps l) and m), in which i∈I is an adjacent index of j, so that there is no k∈I between i and j and also either i=jStart or the reference pair (αi, βi) has already been adapted according to steps l) and m);

[0072] l) determination of j∈J of values h*, k*, l*, m*∈Z with 4$d_{j,1}=\min \left\{\&LeftBracketingBar;{\alpha }_{j,1}\left(h\right)-{\alpha }_i\&RightBracketingBar;+\&LeftBracketingBar;{\beta }_{j,1}\left(k\right)-{\beta }_i\&RightBracketingBar;,\mathrm{with}h,k\in Z\right\}=\&LeftBracketingBar;{\alpha }_{j,1}\left(h^{*}\right)-{\alpha }_i\&RightBracketingBar;+\&LeftBracketingBar;{\beta }_{j,1}\left(k^{*}\right)-{\beta }_i\&RightBracketingBar;;$$d_{j,2}=\min \left\{\&LeftBracketingBar;{\alpha }_{j,2}\left(l\right)-{\alpha }_i\&RightBracketingBar;+\&LeftBracketingBar;{\beta }_{j,2}\left(m\right)-{\beta }_i\&RightBracketingBar;,\mathrm{with}l,m\in Z\right\}=\&LeftBracketingBar;{\alpha }_{j,2}\left(l^{*}\right)-{\alpha }_i\&RightBracketingBar;+\&LeftBracketingBar;{\beta }_{j,2}\left(m^{*}\right)-{\beta }_i\&RightBracketingBar;$$\mathrm{or}\mathrm{for}j\in L\mathrm{of}a\mathrm{value}{k}^{*}\in Z\mathrm{with}$$\&LeftBracketingBar;{\beta }_j\left(k^{*}\right)-{\beta }_i\&RightBracketingBar;=\min \left\{\&LeftBracketingBar;{\beta }_j\left(k\right)-{\beta }_i\&RightBracketingBar;,k\in Z\right\};$

[0073] m) subsequently choice of (αj, βj), in which 5$\mathrm{for}j\in J\left({\alpha }_j,{\beta }_j\right)=\begin{array}{c}\{\left({\alpha }_{j,1}\left(h^{*}\right),{\beta }_{j,1}\left(k^{*}\right)\right),d_{j,1}\le d_{j,2}\\ \{\\ \{\left({\alpha }_{j,2}\left(l^{*}\right),{\beta }_{j,2}\left(m^{*}\right)\right),d_{j,1}>d_{j,2}\end{array}\mathrm{or}\mathrm{for}j\in L\left({\alpha }_j,{\beta }_j\right)=\left({\alpha }_i,{\beta }_j\left(k^{*}\right)\right);$

[0074] for j∈L, αj is in principle randomly selectable.

[0075] The specific choice (αj=αi) is only provisional (until all (αj, βj) are adapted) and guarantees that in step l) there is always a complete reference pair and that freely selectable α angles are ignored (i.e. skipped), so that the spacing from one α angle to the next (adjacent) not freely selectable α angle is minimized.

[0076] When all the pairs (αj,βj) are adapted according to steps l) and m), the freely selectable αj (with j∈L) are no longer required and can be deleted;

[0077] n) choice of a number of triples determined according to steps a) to m) 6$\left({\alpha }_i,{\beta }_i,{\phi }_i\right)=\begin{array}{c}\{\left({\alpha }_i,{\beta }_i,{\phi }_i\right)\mathrm{for}i\in J\\ \{{(}^{*},{\beta }_i,{\phi }_i)\mathrm{for}i\in L\\ \{{(}^{*}{,}^{*},{\phi }_i)\mathrm{for}i\in I,\end{array}$

[0078] in which the components designated with * are randomly selectable;

[0079] o) adapting the selectable components in the triples (αi, βi, φi);

[0080] interpolation of the triples (αi, βi, φi) selected instep n) by a spline and using the interpolation function in q(α,β,φ).

[0081] Interpolated path curves for robot movements determined in this way more particularly avoid the undesired loop formation shown in FIG. 1a, even in the case of subsequent changes to individual components of motion, such as the orientation in space, because such motion components have, according to the invention, their own parametrization in each case, which can be adapted without any negative effects on the other components of motion.

[0082] The adaptation of the freely selectable components of (αi,βi,φi) takes place according to a further development of the above algorithm using:

[0083] iMin:=min {i∈I}

[0084] iMax:=max {i∈I}

[0085] iMin:=min {j∈J}

[0086] jMax:=max {j∈J},

[0087] so that 7${\alpha }_p=\begin{array}{c}\{{\alpha }_{j\mathrm{Min}}\mathrm{for}1\le p<j\mathrm{Min},\\ \{\\ \{{\alpha }_{j\mathrm{Max}}\mathrm{for}j\mathrm{Max}<p\le n\end{array}$$\mathrm{and}$${\beta }_p=\begin{array}{c}\{{\beta }_{i\mathrm{Min}}\mathrm{for}1\le p<i\mathrm{Min},\\ \{\\ \{{\beta }_{j\mathrm{Max}}\mathrm{for}i\mathrm{Max}<p\le n.\end{array}$

[0088] It is also possible to select for adjacent i, j∈J for all i<p<j, i.e. p∈J

αp=αi+(p−i)/(j−i)·(αj−αi)

[0089] and correspondingly for adjacent i, j∈I for all i<p<j, i.e. p∈I

βp=βi+(p−i)/(j−i)·(βj−βi).

[0090] The quaternion-based motion control according to the invention is compactly shown in the following table 1 and in FIG. 5 by means of a flow chart.

TABLE 1
Quaternion algorithm
Action	Calculate orientation with rotary vectors and angles
Calculate	q1	q2 = (1, 0, 0, 0)	q3	. . .	qn − 1	qn = (−1, 0, 0, 0)
quaternions
adapt sign
Calculate	({overscore (v)}1, {overscore (φ)}1)	({overscore (v)}2, {overscore (φ)}2)	({overscore (v)}3, {overscore (φ)}3)	. . .	({overscore (v)}n − 1, {overscore (φ)}n − 1)	({overscore (v)}n, {overscore (φ)}n)
rotary vectors/		with				with {overscore (v)}n = {overscore (v)}n − 1
angles		{overscore (v)}2 = {overscore (v)}1
Minimize	(v1, φ1)	(v2, φ2)	(v3, φ3)	. . .	(vn − 1, φn − 1)	(vn, φn)
changes between
pairs
Delete	(v1, φ1)	(*, φ2)	(v3, φ3)	. . .	(vn − 1, φn − 1)	(*, φn)
freely selectable
vi
Action	Calculate expanded polar coordinates for rotary vectors
Consider	v1 = (0, 0, −1)	*	v3	. . .	vn − 1 = (0, 0, 1)	*
rotary vectors
Calculate	({overscore (α)}1, {overscore (β)}1)	(*, *)	({overscore (α)}3, {overscore (β)}3)	. . .	({overscore (α)}n − 1, {overscore (β)}n − 1)	(*, *)
polar coordinates	with {overscore (α)}1 = {overscore (α)}3				with {overscore (α)}n − 1 = {overscore (α)}n − 2)
Calculate	(α1, β1)	(*, *)	(α3, β3)	. . .	(αn − 1, βn − 1)	(*, *)
expanded
polar coordinates
Delete	(*, β1)	(*, *)	(α3, β3)	. . .	(*, βn − 1)	(*, *)
freely selectable
αi
Action	Represent orientation with “quaternion-polar coordinates”
Calculate	(*, β1, φ1)	(*, *, φ2)	(α3, β3, φ3)	. . .	(*, βn − 1, φn − 1)	(*, *, φn)
three-
dimensional
splines by

[0091] 2. Orientation Control or Guidance with Euler Angles

[0092] Whilst by definition the “classic” Euler angles are restricted to a part of R3 with [−π,π]×[−π/2,π/2]×[−π,π], according to the invention all Euler angle restrictions are dropped. Among all the possible variants of the orientation representation with “expanded Euler angles” from R×R×R (R3) those leading to the minimum reorientations are sought.

[0093] On successively performing three rotations by angles A, B and C, rotating in each case about an axis in the jointly rotated system, said angles are referred to as Euler angles. Independently of the sequence of the axes about which rotation takes place, for each orientation there are twelve different types of Euler angles, e.g. X-Y-Z Euler angles, X-Y-X Euler angles, etc.

[0094] If A, B and C are e.g. Z-Y-X Euler angles, then the total rotation R is obtained from

R =rot(z,A)·Rot(y,B)·Rot(x,C),

[0095] in which Rot(v,ω) designates a rotation about v with angle ω.

[0096] The following considerations can be transferred to all Euler angle types. Thus, without restriction, there can be a limitation to Z-Y-X Euler angles. In this case, the target orientation is obtained from the starting orientation (reference system I), in that

[0097] in reference system I rotation takes place about the Z axis with A (result: reference system I′);

[0098] in the once-primed reference system I′ rotation takes place about the y′ axis with B (result: reference system I″):

[0099] in the twice-primed reference system I″ rotation about the x″ axis with C (result: target orientation R).

[0100] This is shown in exemplified manner in FIG. 2a for Z-Y-X Euler angles. Firstly the reference system I is transformed by a rotation with A=π/4 about the Z axis (positive counterclockwise rotation direction) is transformed into the reference system I′. Subsequently by rotation about the newly defined y′ axis with B=π/3, the reference system I′ is transformed into a new reference system I″. Finally, by rotating about the axis x″ with C=π/4, this passes into the desired target orientation.

[0101] On considering only the x″ axis in the reference system I″ (or in the case of target orientation), the angles A and B clearly represent the spherical coordinates thereof with respect to the starting orientation. The same x″ is obtained with the angle pair (A+π,π−B).

[0102] However, in this case y″ and z″ are rotated by an angle π with respect to the case with (A, B, C). Thus, the same target orientation is obtained when using Euler angles (A+π,π−B,C+π). Thus, in the preceding example, in place of (π/4, π/3, π/4), the trio ({fraction (5/4)}π, ⅔π, {fraction (5/4)}π). This is shown in FIG. 2b).

[0103] According to the invention, such an Euler angle adaptation is brought about in that for minimum reorientation successively the following steps are performed and without restriction of the general nature, the Euler angles A,B,C indicate rotations about the instantaneous z axis, y axis or x axis:

[0104] Representation of the orientations (Ai, Bi, Ci) by:

[0105] a) determination of standard Euler angles (Ai, Bi, Ci),

[0106] b) for I:={i=1, . . . , n|Bi≠±π/2+2kπ, k∈Z} and i∈I definition of

(Ai,1(g), Bi,1(h), Ci,1(k))=(Ai+2gπ, Bi+2hπ, Ci+2kπ), g, h, k∈Z; and

(Ai,2(1), Bi,2(m), Ci,2(o)=(Ai+π+2lπ, π−Bi+2mπ, Ci+π+2oπ), l, m, o∈Z;

[0107] c) for J:={i=1, . . . , n|Bi=π/2+2kπ, k∈Z} and i∈J definition of

(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci+t+2kπ), t∈R, h, k∈Z

[0108] (for i∈J the alternative representation (Ai(t)+π, π−Bi(h), Ci(k, t)+π) supplies no new solutions);

[0109] d) for L:={i=1, . . . , n|Bi=−π/2+2kπ, k∈Z} and i∈L definition of

(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci−t+2kπ), t∈R, h, k∈Z

[0110] (for i∈L the representation (Ai(t)+π, π−Bi(h), Ci(k, t)+π) supplies no new solutions);

[0111] e) choice of a random starting pair (AiStart, BiStart, CiStart) with iStart∈I, e.g. iStart=min {i∈I}; iStart∈J∪L would give a suboptimum orientation control, because the starting pair is itself no longer adapted and therefore the parameter t freely selectable for iStart∈J∪L would have to be predetermined in fixed form;

[0112] f) for all j∈{1, . . . , n) apart from j=iStart determination of expanded Euler angles (Aj, Bj, Cj) according to steps g) and h), where i∈{1, . . . , n} is an adjacent index, i.e. i=j+1 or i=j−1 and additionally either i=iStart or a reference pair (Ai, Bi, Ci) has already been adapted according to steps g) and h):

[0113] g) for j∈I determination of

d j,1 =min{|Aj,1(g)−Ai|+|Bj,1(h)−Bi|+|Cj,1(k)−Ci|, g, h, k∈Z};

d j,2 =min{|Aj,2(l)−Ai|+|Bj,2(m)−Bi|+|Cj,2(o)−Ci|, l, m, o∈Z};

[0114] assuming the minimum for g*, h*, k* or l*, m*, o*; and choice of 8$\left(A_j,B_j,C_j\right)=\begin{array}{c}\{\left(A_{j,1}\left(g^{*}\right),B_{j,1}\left(h^{*}\right),C_{j,1}\left(k^{*}\right)\right),\\ \{\mathrm{if}{d}_{j1}\le d_{j2}\\ \{\\ \{\left(A_{j,2}\left(l^{*}\right),B_{j,2}\left(m^{*}\right),C_{j,2}\left(o^{*}\right)\right),\\ \{\mathrm{if}{d}_{j,1}>d_{j,2};\end{array}$

[0115] h) for j∈J or j∈L definition of

dA j (t):=|Aj(t)−Ai|

dB j (h):=|Bj(h)−Bi|

dC j (k,t):=|Cj(k, t)−Ci|

[0116] with h, k∈Z and t∈R;

[0117] determination of h*∈Z with

dB j (h*)=min{dBj(h), h∈Z};

[0118] calculation of zero points of dAj(t) or dCj(k, t) with respect to t, in which ta0,j is the zero point of dAj(t) and tc0,j(k) is the zero point of dCj(k, t) dependent on the parameter k; choice of k*∈Z, so that there is a minimum spacing of the zero points, i.e.

|ta0,j−tc0,j(k*)|=min{|ta0,j−tc0,j(k)|, k∈Z};

[0119] choice of t* random between ta0,j and tc0,j(k*), because there is always fulfilment there of

dA j (t*)+dBj(h*)+dCj(k*, t*)=min{dAj(t)+dBj(h)+dCj(k, t), t∈R, h, k∈Z}

[0120] and choice of

(Aj, Bj, Cj)=(Aj(t*), Bj(h*), Cj(k*, t*));

[0121] interpolation of the triple (Ai, Bi, Ci) by a spline and use of the interpolation function in the rotation matrix R(A, B, C).

[0122] According to a preferred development of the above-described algorithm, the parameter t* is chosen according to t*=½(ta0,j+tc0,j(k*)). This leads to equal amount orientation changes in the A and C angles.

[0123] The advantage of the above-described algorithm for the determination of expanded Euler angles is illustrated by the following example shown in FIGS. 3 and 4:

[0124] There are two triples of Z-Y-X Euler angles

(A0, B0, C0)=(0, π/2−½, π)

(A1, B1, C1)=(π/2, π/2−½, 0).

[0125] If these values are guided linearly into one another, i.e. by linear interpolation, then for the associated coordinate rotations there is a configuration as shown in FIG. 3.

[0126] FIG. 3 shows how the initial reference system I is transformed by linear interpolation of the angular overall change for each coordinate axis into a target orientation R. As a result of the limitation of the value range for classic Euler angles, in certain circumstances there are undulating movements for individual coordinate axes, as is apparent in FIG. 3 relative to the movement of coordinate axis Z. In the lower area of FIG. 3 reaches the boundary of its definition range and then returns to the smaller value range, which brings about said undulation, which in the same way as the loop formation according to FIG. 1a, represents an uneconomic form of orientation control for robots and which must therefore be avoided.

[0127] If instead (starting from the initial orientation) determination takes place of the adapted Euler trio, then we obtain:

(−π/2, π/2+½, π)=(A1+π−2π, π−B, C1+π).

[0128] As shown in FIG. 4, the associated rotation matrixes consequently pass into one another over the intuitively shortest path, so that the robot movement follows a much more economic course, which has a positive effect regarding the duration and space requirement for manipulation processes. Characteristic undulations are avoided, unlike in the case of FIG. 3.

[0129] The movement control by means of expanded Euler angles according to the invention is illustrated in table 2 and FIG. 6 by means of a flow chart and in compact form.

TABLE 2
Algorithm with expanded Euler angles
Action
	Determine expanded Euler angles
Calculate	({overscore (A)}1, {overscore (B)}1, {overscore (C)}1)	({overscore (A)}2, {overscore (B)}2, {overscore (C)}2)	. . .	({overscore (A)}n−1, {overscore (B)}n−1, {overscore (C)}n−1)	({overscore (A)}n, {overscore (B)}n, {overscore (C)}n)
standard
Euler	{overscore (B)}1 = π/2			{overscore (B)}n−1 = − π/2
angles
Determine	(A1, B1, C1)	(A2, B2, C2)	. . .	(An−1, Bn−1, Cn−1)	(An, Bn, Cn)
expanded
Euler angles from iStart	1	= (Aistart, Bistart, Cistart)	2
	Represent orientation with expanded Euler angles.
Calculate	(A1, B1, C1)	(A2, B2, C2)	. . .	(An−1, Bn−1, Cn−1)	(An, Bn, Cn)
three-
dimensional
spline by

[0130] By means of a block diagram, FIG. 7 shows a possible development of a device according to the invention and which is suitable for controlling the movements of a manipulator in accordance with the above-described method. A manipulator is shown in the form of an industrial robot 1 having an associated control means 2, which according to the invention in particular has computing means 3, which are set up in the form of a compiler/interpreter 4 and interpolation unit and through which the aforementioned algorithms for movement control can be implemented from the control standpoint in that the computing means 3 act within the control means 2 on drive means 6 of robot 1. By means of a manually operable control unit 8, an operator 7 can act on the sequences in the control means 2, specifically in the computing means 3, e.g. by a subsequent programming of reorientations of robot 1.

Claims

1. Method for the controlling movements of a manipulator, such as an industrial robot, given point sequences of poses (positions and orientations) being interpolated by splines, wherein components of the motion are separately parametrized.

2. Method according to claim 1, wherein positions of the manipulator are parametrized separately from the orientations thereof.

3. Method according to claim 1, wherein movements of auxiliary axes are parametrized separately from the positions and/or orientations of the manipulator.

4. Method according to claim 1, wherein an orientation control takes place by means of quaternions (qi, i=1, . . . , n).

5. Method according to claim 4, wherein for minimum reorientation between individual points successively the following steps are performed: representation of each quaternion on the four-dimen sional unit sphere according toq(α,β,φ)=(cos ½φ), sin(½φ)·v(α, β))with(v(α,β)=(cos(α) cos(β), sin(α) cos(β), sin(β)),in which α, β, (and r=1) give the spherical coordinates of the rotating vectors v=v(α, β) determination of motion-optimum pairs (vi, φi) by a) adapting the sign of qi, so thatqiTqi+1≧0 for all i=1, . . . , n−1;b) for all i=1,b) for all i=1 . . . , n determination of a possible pair (vi, φi) withqi=(cos(½φi), sin(½φi)vi);c) choice of a random starting pair (ViStart, φiStart), with iStart∈I:={i=1, . . . ,n|qi≠(±1, 0, 0, 0)}, e.g. iStart=min{i∈I}; on choosing iStart∈I, the completely freely selectable rotary vector viStart would be unnecessarily fixed, which can lead to a suboptimum orientation, because the starting pair would itself no longer be adapted; d) for all j∈{1, . . . , n} with the exception of j=iStart determination of in each case precisely one pair (vj, φj) starting from a reference pair (vi, φi) in accordance with steps e) to g), in which i∈{1, . . . , n) is a neighbouring index, i.e. i=j+1 or i=j−1 and also either i=iStart or the reference pair has already been adapted according to steps e) to g); e) determination of k*, h*∈Z, so that 9&LeftBracketingBar;φj,1⁡(k*)-φi&RightBracketingBar;=min⁢{&LeftBracketingBar;φj,1⁡(k)-φi&RightBracketingBar;,k∈Z}=min⁢{&LeftBracketingBar;φj+4⁢k⁢ ⁢π-φi&RightBracketingBar;,k∈Z},&LeftBracketingBar;φj,2⁡(h*)-φi&RightBracketingBar;=min⁢{&LeftBracketingBar;φj,2⁡(h)-φi&RightBracketingBar;,h∈Z}=min⁢{&LeftBracketingBar;-φj+4⁢h⁢ ⁢π-φ⁢ ⁢i&RightBracketingBar;,h∈Z};f) subsequently for j∈I determination ofdj,1=|arccos(viTvj)|+|(φj,1(k*)−φi|,dj,2=|arccos(viT(−vj))|+|(φj,2(h*)−φi|or for j∉Idj,1=|φj,1(k*)−φi|,dj,2=|φj,2(h*)−φi|;g) subsequently in each case choice of (vj, φj), where for 10j∈I⁢(vj,φj)={(vj,φj,1⁡(k*)),dj,1≤dj,2{{(-vj,φj,2⁡(h*)),dj,1>dj,2or⁢ ⁢for⁢ ⁢j∈I⁢(vj,φj)={(vi,φj,1⁡(k*)),if⁢ ⁢dj,1≤dj,2{{(vi,φj,2⁡(h*)),if⁢ ⁢dj,1>dj,2;determination of pairs (αi, βi) from vi in spherical coordinate representation by h) calculation of standard spherical coordinates (αi, βi) of vi for all i∈I:vi(cos(αi) cos(βi), sin(αi) cos(βi), sin(βi))i) definition ofJ:={j∈I with vj≠(0, 0, ±1)},L:={j∈I with vj=(0, 0, ±1)}and therefore definition of expanded spherical coordinates from R×R for i∈J:(αi,1(h), βi,1(k))=(αi+2hπ, βi+2kπ), h, k∈Z;(αi,2(1), βi,2(m))=(αi+π+2lπ, π−βi+2mπ), l, m∈Z;for i∈L:(αi(x), βi(k)) (x, Bi+2kπ), x∈R, k∈Z;j) fixing a random starting pair (αjStart, βjStart) with jStart∈J; k) for all j∈I with the exception of j=jStart in each case determination of expanded polar coordinates in accordance with steps l) and m), in which i∈I is an adjacent index of j, so that there is no k∈I between i and j and also either i=jStart or the reference pair (αi,βi) has already been adapted in accordance with steps l) and m): l) determination of j∈J of values h*, k*, l*, m*∈Z with 11dj,1⁢=min⁢{&LeftBracketingBar;αj,1⁡(h)-αi&RightBracketingBar;+&LeftBracketingBar;βj,1⁡(k)-βi&RightBracketingBar;,with⁢ ⁢h,k∈Z}⁢=&LeftBracketingBar;αj,1⁡(h*)-αi&RightBracketingBar;+&LeftBracketingBar;βj,1⁡(k*)-βi&RightBracketingBar;;dj,2⁢=min⁢{&LeftBracketingBar;αj,2⁡(l)-αi&RightBracketingBar;+&LeftBracketingBar;βj,2⁡(m)-βi&RightBracketingBar;,with⁢ ⁢l,m∈Z}⁢=&LeftBracketingBar;αj,2⁡(l*)-αi&RightBracketingBar;+&LeftBracketingBar;βj,2⁡(m*)-βi&RightBracketingBar;or⁢ ⁢for⁢ ⁢j∈L⁢ ⁢of⁢ ⁢a⁢ ⁢value⁢ ⁢k*∈Z⁢ ⁢with&LeftBracketingBar;βj⁡(k*)-βi&RightBracketingBar;=min⁢{&LeftBracketingBar;βj⁡(k)-βi&RightBracketingBar;,k∈Z};m) subsequently choice of (αj, βj), in which for j∈J 12(αj,βj)={(αj,1⁡(h*),βj,1⁡(k*))⁢ ⁢dj,1≤dj,2{{(αj,2⁡(l*),βj,2⁡(m*))⁢ ⁢dj,1>dj,2or⁢ ⁢for⁢ ⁢j∈L(αj,βj)=(αi,βj⁡(k*));n) choice of a number of triples determined according to steps a) to m) 13(αi,βi,φi)={(αi,βi,φi)⁢ ⁢for⁢ ⁢i∈J{(*⁢,βi,φi)⁢ ⁢for⁢ ⁢i∈ ⁢L{(*⁢,*,φi)⁢ ⁢for⁢ ⁢i∈ ⁢I,in which all components designated * can be selected at random; o) adaptation of the selectable components in the triples (αi, βi, φi); interpolation of the triples (αi, βi, φi) selected in step n) by a spline; and use of the interpolation function in q(α,β,φ).

6. Method according to claim 5, wherein adaptation takes place of the freely selectable components of (αi, βi, φi) using: iMin:=min {i∈I}iMax:=max {i∈I}iMin:=min {j∈J}jMax:=max {j∈J}, so that 14αp={(αjMin⁢ ⁢for⁢ ⁢1≤p<jMin,{{αjMax⁢ ⁢for⁢ ⁢jMax<p≤nandβp={(βiMin⁢ ⁢for⁢ ⁢1≤p<iMin,{{βiMax⁢ ⁢for⁢ ⁢iMax<p≤n.

7. Method according to claim 6, wherein for adjacent i, j∈J for all i<p<j, i.e. p∉J a choice is made of

8. Method according to claim 1, wherein an orientation control takes place by means of successive rotations through Euler angles A1, Bi, Ci, i=1, . . . , n, and in each case directly succeeding rotations take place about axes perpendicular to one another.

9. Method according to claim 8, wherein use is made of expanded Euler angles Ai, Bi, Ci∈R.

10. Method according to claim 9, wherein for minimum reorientation successively the following steps are performed and the Euler angles A, B and C, without restricting the general nature, indicate rotations about the instantaneous z axis, y axis or x axis: representation of the orientations (Ai, Bi, Ci) by: a) determination of standard Euler angles (Ai, Bi, Ci) b) for I:={i=1, . . . , n|Bi≠π/2+2kπ, k∈Z} and i∈I definition of(Ai,1(g), Bi,1(h), Ci,1(k))=(Ai+2gπ, Bi+2hπ, Ci+2kπ), g, h, k∈Z; and(Ai,2(1), Bi,2(m), Ci,2(o))=(Ai+π+2lπ, π−Bi+2mπ, Ci+π+2oπ), l, m, o∈Z;c) for J:={i=1, . . . , n|Bi=π/2+2kπ, k∈Z} and i∈J definition of(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci+t+2kπ), t∈R, h, k∈Zd) for L:={i=1, . . . , n|Bi=π/2+2kπ, k∈Z} and i∈L definition of(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci−t+2kπ), t∈R, h, k∈Ze) choice of a random starting pair (AiStart, BiStart, CiStart) with iStart∈I, e.g. iStart=min {i∈I}; f) for all j∈{1, . . . , n) apart from j=iStart determination of expanded Euler angles (Aj, Bj, Cj) according to steps g) and h), where i∈{1, . . . ,n} is an adjacent index, i.e. i=j+1 or i=j−1 and additionally either i=iStart or a reference pair (Ai, Bi, Ci) has already been adapted according to steps g) and h): g) for j∈I determination ofdj,1=min{|Aj,1(g)−Ai|+|Bj,1(h)−Bi|+|Cj,1(k)−Ci|, g, h, k∈Z};dj,2=min{|Aj,2(l)−Ai|+|Bj,2(m)−Bi|+|Cj,2(o)−Ci|, l, m, o∈Z};assuming the minimum for g*, h*, k* or l*, m*, o*; and choice of 15(Aj,Bj,Cj)={(Aj,1⁡(g*),Bj,1⁡(h*),Cj,1⁡(k*)),{if⁢ ⁢dj,1≤dj,2{{(Aj,2⁡(l*),Bj,2⁡(m*),Cj,2⁡(o*)),{if⁢ ⁢dj,1>dj,2;h) for j∈J or j∈L definition ofdAj(t):=|Aj(t)−Ai|dBj(h):=|Bj(h)−Bi|dCj(k, t):=|Cj(k, t)−Ci|with h, k∈Z and t∈R; determination of h*∈Z withdBj(h*)=min{dBj(h), h∈Z};calculation of zero points of dAj(t) or dCj(k, t) with respect to t, in which ta0,j is the zero point of dAj(t) and tc0,j(k) is the zero point of dCj(k, t) dependent on the parameter k; choice of k*∈Z, so that there is a minimum spacing of the zero points, i.e.|ta0,j−tc0,j(k*)|=min{|ta0,j−tc0,j(k)|, k∈Z};choice of t* random between ta0,j and tc0,j(k*), because there is always fulfilment there ofdAj(t*)+dBj(h*)+dCj(k*, t*)=min{dAj(t)+dBj(h)+dCj(k, t), t∈R, h, k∈Z}and choice of(Aj, Bj, Cj)=(Aj(t*), Bj(h*), Cj(k*, t*));interpolation of the triple (Ai, Bi, Ci) by a spline and use of the interpolation function in the rotation matrix R(A, B, C).

11. Method according to claim 10, wherein t*=½(ta0,j+tc0,j(k*)) is chosen.

12. Device for controlling the movements of a manipulator, such as a robot, having an interpolation unit for interpolating a given point sequence of poses (positions and orientations) by splines, wherein the interpolation unit is constructed for separate parametrization of components of motion during interpolation.

13. Device according to claim 12, wherein the interpolation unit is constructed for the separate parametrization of positions and orientations of the manipulator.

14. Device according to claim 12, wherein the interpolation unit is constructed for the separate parametrization of on the one hand movements of auxiliary axes and on the other positions and/or orientations of the manipulator.

15. Device according to claim 12, wherein the interpolation unit is set up for an orientation control by means of quaternions (qi, i=1, . . . , n).

16. Device according to claim 15, wherein the interpolation unit is set up for performing the following steps: representation of each quaternion on the four-dimensional unit sphere according toq(α,β,φ)=(cos(½φ), sin(½φ)·v(α,β))withv(α,β)=(cos(α) cos(β), sin(α) cos(β), sin(β)),in which α,β (and r=1) indicate the spherical coordinates of the rotary vectors v=v(α,β); determining motion-optimum pairs (vi, φi) by a) adapting the sign of qi, so thatqiTqi+1>0 for all i=1, . . . , n−1;b) for all i=1, . . . , n determination of a possible pair (vi, φi) withqi=(cos(½φi), sin(½φi)vi);c) choice of a random starting pair (viStart, φiStart) with iStart∈I:={i=1, . . . , n|qi≠(±1, 0, 0, 0)}; d) for all j∈{1, . . . , n) with the exception of j=iStart determination of in each case precisely one pair (vj, φj) starting from a reference pair (vi, φi) in accordance with steps e) to g), in which i∈{1, . . . , n} is an adjacent index, i.e. i=j+1 or i=j−1 and also either i=iStart or the reference pair has already been adapted according to steps e) to g); e) determination of k*, h*∈Z, so that 16&LeftBracketingBar;φj,1⁡(k*)-φi&RightBracketingBar;=⁢min⁢{&LeftBracketingBar;φj,1⁡(k)-φi&RightBracketingBar;,k∈Z} ⁢=⁢min⁢{&LeftBracketingBar;φj+4⁢k-φi&RightBracketingBar;,k∈Z},&LeftBracketingBar;φj,2⁡(h*)-φi&RightBracketingBar;=⁢min⁢{&LeftBracketingBar;φj,2⁡(h)-φi&RightBracketingBar;,h∈Z} ⁢=⁢min⁢{&LeftBracketingBar;-φj+4⁢h⁢ ⁢π-φi&RightBracketingBar;,h∈Z};f) subsequently for j∈I determination ofdj,1=|arccos(viTvj)|+|φj,1(k*)−φi|,dj,2=|arccos(viT(−v))|+|φj,2(h*)−φi|or for j∉Idj,1=|φj,1(k*)−φi|,dj,2=|φj,2(h*)−φi|;g) subsequently in each case choice of (vj, φj), where for j ε I 17(vj,φj)={(vj,φj,1⁡(k*))⁢ , ⁢dj,1≤dj,2{{(-vj,φj,2⁡(h*))⁢ , ⁢dj,1>dj,2or⁢ ⁢for⁢ ⁢j∉I(vj,φj)={(vi,φj,1⁡(k*))⁢ ⁢if⁢ ⁢dj,1≤dj,2{{(vi,φj,2⁡(h*))⁢ ⁢if⁢ ⁢dj,1>dj,2;determination of pairs (αi, βi) from vi in spherical coordinate representation by: h) calculating standard spherical coordinates (αi, βi) of vi for all i∈I:vi=(cos(αi) cos(βi), sin(αi) cos(βi), sin(βi));i) definition ofJ:={j∈I with vj≠(0, 0, ±1)},L:={j∈I with vj=(0, 0, ±1)}and therefore definition of expanded spherical coordinates from R×R for i∈J:(αi,1(h), βi,1(k)=(αi+2hπ, βi+2kπ), h, k∈Z;(αi,2(l), βi,2(m))=(αi+π+2lπ, π−βi+2mπ),l, m∈Z;for i∈L:(αi(x), βi(k))=(x, βi+2kπ), x∈R, k∈Z;j) fixing a random starting pair (αjStart, βiStart) with jStart∈J; k) for all j∈I with the exception of j=jStart, in each case single determination of expanded polar coordinates according to steps l) and m), in which i∈I is an adjacent index of j, so that there is no k∈I between i and j and also either i=jStart or the reference pair (αi, βi) was already adapted according to steps l) and m); l) determination for j∈J of values h*, k*, l*, m*∈Z with 18dj,1=⁢min⁢{&LeftBracketingBar;αj,1⁡(h)-αi&RightBracketingBar;+&LeftBracketingBar;βj,1⁡(k)-βi&RightBracketingBar;,with⁢ ⁢h,k∈Z}=⁢&LeftBracketingBar;αj,1⁡(h*)-αi&RightBracketingBar;+&LeftBracketingBar;βj,1⁡(k*)-βi&RightBracketingBar;;dj,2=min⁢{&LeftBracketingBar;αj,2⁡(l)-αi&RightBracketingBar;+&LeftBracketingBar;βj,2⁡(m)-βi&RightBracketingBar;,with⁢ ⁢l,m∈Z} ⁢=⁢&LeftBracketingBar;αj,2⁡(l*)-αi&RightBracketingBar;+&LeftBracketingBar;βj,2⁡(m*)-βi&RightBracketingBar;or⁢ ⁢for⁢ ⁢j∈L⁢ ⁢of⁢ ⁢a⁢ ⁢value⁢ ⁢k*∈Z⁢ ⁢with&LeftBracketingBar;βj⁡(k*)-βi&RightBracketingBar;=min⁢{&LeftBracketingBar;βj⁡(k)-βi&RightBracketingBar;,k∈Z};m)⁢ ⁢subsequent⁢ ⁢choice⁢ ⁢of⁢ ⁢(αj,βj),in⁢ ⁢which⁢ ⁢for⁢ ⁢j∈J(αj,βj)={(αj,1⁡(h*),βj,1⁡(k*))⁢ , ⁢dj,1≤dj,2{{(αj,2⁡(l*),βj,2⁡(m*))⁢ , ⁢dj,1>dj,2or⁢ ⁢for⁢ ⁢j∈L(αj,βj)=(αi,βj⁡(k*));n) choice of a number of triples determined according to steps a) to m) 19(αi,βi,φi)={(αi,βi,φi)⁢ ⁢for⁢ ⁢i∈J{(*⁢,βi,φi)⁢ ⁢for⁢ ⁢i∈ ⁢L{(*⁢,*,φi)⁢ ⁢for⁢ ⁢i∉ ⁢I,where all the components designated * are randomly selectable; o) adapting the selectable components in the trios (αi, βi, φi); interpolation of the triples (αi, βi, φi) selected in step n) by a spline and use of the interpolation function in q(α, β, φ).

17. Device according to claim 16, wherein the interpolation unit is set up for adapting the freely selectable components of (αi, βi, φi) using: iMin:=min {i∈e}iMax:=max {i∈I}iMin:=min {j∈J}jMax:=max {j∈J}so that 20αp={αjMin⁢ ⁢for⁢ ⁢1≤p<jMin,{{αjMax⁢ ⁢for⁢ ⁢jMax<p≤nandβp={(βiMin⁢ ⁢for⁢ ⁢1≤p<iMin,{{βiMax⁢ ⁢for⁢ ⁢iMax<p≤n.

18. Device according to claim 17, wherein the interpolation unit is set up for adjacent i, j∈J for all i<p<j, i.e. p∉J for the determination of

19. Device according to claim 12, wherein the interpolation device is set up for an orientation control by means of successive rotations through Euler angles Ai, Bi, Ci and in each case directly succeeding rotations take place about axes perpendicular to one another.

20. Device according to claim 19, wherein the interpolation unit is set up for the use of expanded Euler angles Ai, Bi, Ci∈R.

21. Device according to claim 20, wherein the interpolation device is set up for the sequential performance of the following steps: representation of the orientations (Ai, Bi, Ci) by: a) determining the standard Euler angles (Ai, Bi, Ci); b) for I:=({i=1, . . . , n|Bi≠π/2+2kπ, k∈Z} and i∈I definition of(Ai,1(g), Bi,1(h), Ci,1(k)=(Ai+2gπ, Bi+2hπ, Ci+2kπ), g, h, k∈Z; and(Ai,2(l), Bi,2(m), Ci,2(o))=(Ai+π+2lπ, π−Bi+2mπ, Ci+π+2oπ), l, m,∈Z;c) for J:={i=1, . . . , n|Bi=π/2+2kπ, k∈Z} and i∈J definition of(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci+t+2kπ), t∈R, h, k∈Z;d) for L:={i=1, . . . , n|Bi=−π/2+2kπ, k∈Z} and i∈L definition of(Ai(t), Bi(h), Ci(k, t))=(Ai+t, Bi+2hπ, Ci−t+2kπ), t∈R, h, k∈Z;e) choice of a random starting pair (AiStart, BiStart, CiStart) with iStart∈I; f) for all j∈{1, . . . , n}, apart from j=iStart, determining expanded Euler angles (Aj, Bj, Cj) according to steps g) and h), in which i∈{1, . . . , n} is an adjacent index, i.e. i=j+1 or i=j−1 and additionally either i=iStart or a reference pair (Ai, Bi, Ci) has already been adapted according to steps g) and h); g) for j∈I determination ofdj,1=min{|Aj,1(g)−Ai|+|Bj,1(h)−Bi|+|Cj,1(k)−Ci|, g, h, k∈Z};dj,2=min{|Aj,2(l)−Ai|+|Bj,2(m)−Bi|+|Cj,2(o)−Ci|, l, m, o∈Z},assuming the minimum for g*, h*, k* or l*, m*, o*; and choice of 21(Aj,Bj,Cj)=⁢{(Aj,1⁡(g*),Bj,1⁡(n*),Cj,1⁡(k*)),⁢{if⁢ ⁢dj,1≤dj,2{{(Aj,2⁡(l*),Bj,2⁡(m*),Cj,2⁡(o*)),{if⁢ ⁢dj,1>dj,2;h) for j∈J or j∈L definition ofdAj(t):=|Aj(t)−Ai|dBj(h):=|Bj(h)−Bi|dCj(k,t):=|Cj(k, t)−Ci|with h, k∈Z and t∈R; determination of h*∈Z withdBj(h*)=min{dBj(h), h∈Z};determination of zero points of dAj(t) or dCj(k, t) with respect to t, in which ta0,j is the zero point of dAj(t) and tc0,j(k) is the zero point of dCj(k, t); choice of k*∈Z, so that the spacing of the zero points is at a minimum, i.e.|ta0,j−tc0,j(k*)|=min{|ta0,j(k)|, k∈Z};choice of t* random between ta0,j and tc0,j(k*); and choice of(Aj, Bj, Cj)=(Aj(t*), Bj(h*), Cj(k*, t*));interpolation of the triples (Ai, Bi, Ci) by a spline and use of the interpolation function in the rotation matrix R (A, B, C).

22. Device according to claim 21, wherein the interpolation unit is set up for the choice of t*=½(ta0,j+tc0,j(k*)).