Skip to main content
U.S. flag

An official website of the United States government

Official websites use .gov
A .gov website belongs to an official government organization in the United States.

Secure .gov websites use HTTPS
A lock ( ) or https:// means you’ve safely connected to the .gov website. Share sensitive information only on official, secure websites.

Pose Mathematics for Path Planning

Pose Mathematics for Path Planning

POSEMATH: Pose Mathematics for Path Planning

Code Organization

Naming Conventions

Usage

Units

Translational Representations

Rotational Representations

Combined Representations

Functions and Operators

Translational Functions

Rotational Functions

Combined Functions

Automatic Conversions in C++

Appendix A: Platform Support and Compilation

NIST-Specific Compilation

Appendix B: Programming Reference

Karl Murphy's Posemath Examples

POSEMATH: Pose Mathematics for Path Planning

POSEMATH is a library of software for representing and manipulating locations in three-dimensional space. POSEMATH can be used by either C or C++ programmers to define coordinate systems and the position and orientation of frames within coordinate systems, and to manipulate these frames to plan manipulator paths.

Code Organization

The POSEMATH software consists of three modules: the pose math code proper, support for printed output and diagnostics, and a sine-cosine utility which enables using the single-instruction sine and cosine on some platforms for efficiency. Each module consists of a code (.c) and header (.h) file. The code is written so that it will compile in both the C and C++ languages, using the ANSI C++ built-in compiler symbol __cplusplus for selection. Debug output can be enabled with the PM_DEBUG symbol. Defining this as a compiler switch will enable diagnostic printing messages when errors are encountered. Leaving it undefined will disable diagnostic printing messages.

The code is compiled into four separate libraries, for C or C++, and with debug output enabled or disabled. Here are the details:

_posemath.c C implementation of pose math functions

posemath.cc C++ implementation of pose math functions

posemath.h Declarations for pose math code, for C and C++

_mathprnt.c Printed output code, for C

mathprnt.cc Printed output code, for C++

mathprnt.h Declarations for printed output code, for C and C++

sincos.c C implementation of processor-specific single-instruction sine and cosine functions

sincos.h Declarations for single-instruction sine and cosine, for C and C++

For Unix and related operating systems, compiled libraries are:

libpm.a All code compiled for C or C++ linkage, no diagnostics

libpmdb.a All code compiled for C or C++ linkage, diagnostics included (PM_PRINT_ERROR defined)

For Microsoft Windows and related operating systems, compiled libraries are:

libpm.a All code compiled for C or C++ static linkage, no diagnostics

pm.dll All code compiled for C or C++ dynamic linkage, no diagnostics

Note that there is no diagnostics support for Microsoft Windows applications, as there is no traditional console available for these messages.

Programmers need only include posemath.h in their code to use the basic POSEMATH library. If printed output is desired, programmers should include mathprnt.h. This file declares the C++ iostream operators for output, and the C counterparts to the printf() functions.

Naming Conventions

C data types in POSEMATH are prefixed with Pm, and follow the case-change convention where the subsequent words in the descriptive name are concatenated with leading capitals, e.g., PmRotationVector. C++ data types are prefixed with PM_ , and are all capital letters with perhaps some additional underscores, e.g., PM_ROTATION_VECTOR. In C, they are declared as structure types with typedef struct. In C++, they are declared as classes. C functions begin with pm, and are followed by the POSEMATH C abbreviations for each type. For example,

PmCartesian the C Cartesian data type

PM_CARTESIAN the C++ Cartesian data type

pmCartCartCross() C function for cross product of two PmCartesian types

C++ versions of the C functions are implemented using operator overloading as much as possible. Where functions are required or more intuitive, these are implemented as overloaded functions and hence do not require a unique prefix. For example,

norm() C++ normalization function for all C++ types

isNorm() C++ normalization predicate for all C++ types

 

Usage

POSEMATH is based in C, with C++ added to make coding easier and more intuitive. Functions that result in a data type take pointers to the result as their last argument, and return an integer error code. For example, the function that computes the cross product of two Cartesian vectors is declared as

int pmCartCartCross(PmCartesian v1, PmCartesian v2, PmCartesian * vout);

C programmers must supply a pointer to existing storage for a PmCartesian for the result of the cross product of v1 and v2.

For some functions, such as the predicate pmCartCartCompare, the return value is the only result, and no pointer to results need be passed.

For C++, additional syntactic interfaces make programs easier to read. This is accomplished through overloading of functions and operators. For example, the multiplication operator * is overloaded to take a scalar and a Cartesian vector, and results in a scaled Cartesian vector. In C, this looks like:

PmCartesian a, b;

double s;

int error;

error = pmCartScalMult(a, s, &b);

The error code can be checked to see if the result is valid (although in this case scalar-vector multiplies are always valid). In C++, the code looks like:

PM_CARTESIAN a, b;

double s;

b = a * s;

which is much more intuitive. However, since the return value is the vector itself, the programmer cannot check the result for validity as before. POSEMATH provides a global integer variable, pmErrno, which is set to 0 when the result of a function or operation is valid, and to a non-zero error code if the function or operation is invalid. C++ programmers using the compact operator syntax should check pmErrno in place of the integer return code in the corresponding function.

The integer return code from the C functions and pmErrno are identical, and are declared in the header file posemath.h. These codes include:

PM_ERR unspecified error

PM_IMPL_ERR function not implemented

PM_NORM_ERR argument should have been normalized

PM_DIV_ERR divide by zero attempted

The use of a global variable pmErrno means that the C++ code is not thread-safe or reentrant. If this is a requirement, then the C functions must be used. This does not mean that the C language needs to be used: the C functions can be called from C++ code, but the programmer loses the syntactic compactness resulting from function- and operator overloading.

Units

Units are immaterial for translation quantities, and are assumed to be radians for angular quantities.

Translational Representations

In our three-dimensional universe, the location of a point in space is uniquely determined by three numbers. These numbers usually represent the coordinates of the point in each of three perpendicular axes, relative to some known origin. The numbers can be thought of as the distance along each axis that the point traversed (or translated) as it moved from the origin to its location.

Translation is often referred to as position. Common translational representations include Cartesian coordinates (X, Y, and Z) and spherical coordinates (azimuth, altitude, and range).

POSEMATH supports three translational representations: Cartesian, cylindrical, and spherical. These are defined as:

C struct C++ class fields C abbreviation
PmCartesian PM_CARTESIAN x, y, z Cart
PmCylindrical PM_CYLINDRICAL theta, r, z Cyl
PmSpherical PM_SPHERICAL theta, phi, r Sph

Fields are of type double. Values for the fields of each type are unconstrained. For example, the values for the angular fields is not required to lie in the range 0 to 2π, or -π to π. However, if these types are returned or result from calculations, they may be normalized to lie within the range [0, 2π) using the rotational equivalence that 0 equals 2π, π equals 3π, etc.

All translational representations are equivalent, in the sense that a location in space expressed in one representation can be converted to any of the other representations. The numbers will in general be different, but the location they represent is the same. POSEMATH provides functions to convert between these representations. Using the POSEMATH naming conventions, these function declarations are:

int pmCartCylConvert(PmCartesian, PmCylindrical *);

int pmCartSphConvert(PmCartesian, PmSpherical *);

int pmCylCartConvert(PmCylindrical, PmCartesian *);

int pmCylSphConvert(PmCylindrical, PmSpherical *);

int pmSphCartConvert(PmSpherical, PmCartesian *);

int pmSphCylConvert(PmSpherical, PmCylindrical *);

The first argument is the type to be converted from. The second argument is a pointer to the object to be converted to. The integer return value is 0 if successful, PM_NORM_ERR if unsuccessful, or PM_IMPL_ERR if unimplemented. For example, converting from a PmCartesian v to a PmSpherical s looks like:

pmCartCylConvert(v, &s);

For C++ programmers, converting between types can be done using the = operator, which has been overloaded to call these functions. Extending the above example, converting from a Cartesian to spherical coordinate representation in C++ looks like:

s = v;

Since the return value of the = operator is the converted type, the programmer should check the value of pmErrno for proper conversion if desired.

Rotational Representations

In addition to a location in space, an object has an orientation. Like location, orientation requires three numbers to be uniquely specified. The numbers can be thought of as the angles that the object is inclined to with respect to some reference planes.

Rotational representations often use more than three numbers to specify orientation. This may be to make the representation more intuitive, or for computational efficiency.

POSEMATH supports six rotational representations: rotation vectors, rotation matrices, quaternions, ZYZ Euler angles, ZYX Euler angles, and roll-pitch-yaw. These are defined as:

C struct C++ class fields C abbreviation
PmRotationVector PM_ROTATION_VECTOR s, x, y, z Rot
PmRotationMatrix PM_ROTATION_MATRIX PmCartesian x, y, z Mat
PmQuaternion PM_QUATERNION s, x, y, z Quat
PmEulerZyz PM_EULER_ZYZ z, y, zp Zyz
PmEulerZyx PM_EULER_ZYX z, y, x Zyx
PmRpy PM_RPY r, p, y Rpy

Fields are of type double, except for a PmRotationMatrix, whose fields are of type PmCartesian. Values for the fields of each type may be constrained. For example, each PmCartesian element of a PmRotationMatrix must have a magnitude of 1.0 (each must be a unit vector), and they must be mutually perpendicular. Normalization functions are provided which take any of these rotational types as their first argument, a pointer to the same rotational type as their second argument, and place the normalization of the first into the second. PM_NORM_ERR is returned if the normalization could not take place. This will occur if the first argument is entirely 0. For example, normalizing a PmQuaternion q looks like:

PmQuaternion q;

pmQuatNorm(q, &q);

C++ programmers can use the overloaded function norm() to accomplish the same thing in a cleaner way:

PM_QUATERNION q;

q = norm(q);

Since the error code is not returned directly, C++ programmer should check pmErrno to make sure the conversion was successful.

All rotational representations are equivalent, in the sense that a rotation in space expressed in one representation can be converted to any of the other representations. The numbers will in general be different, but the rotation they represent is the same. POSEMATH provides functions to convert between these representations. Using the POSEMATH naming conventions, these function declarations are:

int pmRotMatConvert(PmRotationVector, PmRotationMatrix *);

int pmQuatZyzConvert(PmQuaternion, PmEulerZyz *);

int pmZyxRpyConvert(PmEulerZyz, PmRpy *);

(Not all of the functions are listed). The first argument is the type to be converted from. The second argument is a pointer to the object to be converted to. The integer return value is 0 if successful, PM_NORM_ERR if unsuccessful, or PM_IMPL_ERR if the function is unimplemented. For example, converting from a PmQuaternion q to a PmRotationMatrix m looks like:

pmQuatMatConvert(q, &m);

For C++ programmers, converting between types can be done using the = operator, which has been overloaded to call these functions. Extending the above example, converting from a PM_QUATERNION q to a PM_ROTATION_MATRIX m in C++ looks like:

m = q;

Since the return value of the = operator is the converted type, the programmer should check the value of pmErrno for proper conversion if desired.

Axis-Angle Representations

Another rotational type, the axis-angle, is provided for representing rotations about one of the principal Cartesian axes by a given angle. This type, PmAxis, is the same for both C and C++ and may take the values PM_X, PM_Y, and PM_Z only, which represent rotations about the X, Y, and Z Cartesian axes, respectively. These functions are provided:

int pmAxisAngleQuatConvert(PmAxis, double, PmQuaternion *);

int pmQuatAxisAngleMult(PmQuaternion, PmAxis,

double, PmQuaternion *);

The first converts a rotation about the given axis by the given angle to a quaternion. The second computes the rotation resulting from an initial quaterion in the base frame, then a rotation by the given axis-angle in the quaternion frame.

Combined Representations

The complete representation of an objectÆs location and orientation in space requires both translational and rotational representations. Of the many combinations of translational and rotational types possible, the two most common have been defined. These are the homogeneous transform and the pose.

The homogeneous transform uses a Cartesian translation representation and a rotation matrix for rotation. The pose uses a Cartesian translation representation and a quaternion. These are defined as:

C struct C++ class fields C abbreviation
PmHomogeneous PM_HOMOGENEOUS PmCartesian tran, PmRotationMatrix rot Hom
PmPose PM_POSE PmCartesian tran,PmQuaternion rot Pose

Since these combined representations include elements which require normalization, normalization functions are provided which take any of these combined types as their first argument, a pointer to the same combined type as their second argument, and place the normalization of the first into the second. PM_NORM_ERR is returned if the normalization could not take place. For example, normalizing a PmPose p looks like:

pmPoseNorm(p, &p);

C++ programmers can use the overloaded function norm() to accomplish the same thing in a cleaner way:

PM_POSE p;

p = norm(p);

Since the error code is not returned directly, C++ programmer should check pmErrno to make sure the conversion was successful.

All combined representations are equivalent, in the sense that a position and orientation in space expressed in one representation can be converted to any of the other representations. The numbers will in general be different, but the position and orientation they represent is the same. POSEMATH provides functions to convert between these representations. Using the POSEMATH naming conventions, these function declarations are:

int pmPoseHomConvert(PmPose, PmHomogeneous *);

int pmHomPoseConvert(PmHomogeneous, PmPose *);

The first argument is the type to be converted from. The second argument is a pointer to the object to be converted to. The integer return value is 0 if successful, PM_NORM_ERR if unsuccessful, or PM_IMPL_ERR if the function is unimplemented. For example, converting from a PmPose p to a PmHomogeneous h looks like:

pmPoseHomConvert(p, &h);

For C++ programmers, converting between types can be done using the = operator, which has been overloaded to call these functions. Extending the above example, converting from a PM_POSE p to a PM_HOMOGENEOUS h in C++ looks like:

h = p;

Since the return value of the = operator is the converted type, the programmer should check the value of pmErrno for proper conversion if desired.

Functions and Operators

Mathematical operations for translational and rotational types are provided to carry out calculations typical of manipulator path planning in three-space. In general, functions that work on a particular representation will also work on other equivalent representations.

Detailed descriptions of each function are found in Appendix B.

Translational Functions

The translational data representations are used to represent position vectors, and all the typical vector operations have been provided. These include comparison, the dot product, cross product, norm (unit vector), magnitude, sum, difference, magnitude of difference (displacement), scalar multiply and divide, and generalized inverse.

For C functions that take two arguments, the arguments must be the same type. For C++ overloaded functions or operators that take two arguments, they may be of different type. In this case, the constructors are called to do the conversions automatically.

Return values for all the C functions are integer error codes, with the exception of the predicate functions which return 1 if it is true or 0 if not.

Return values for all the C++ functions are the result that corresponds to the last argument of the C functions. The global variable pmErrno should be checked to see if the function was successful.

Following the POSEMATH naming conventions, these functions are prefixed with pm, the abbreviated names of their arguments, and the operation. In the table below, X indicates any of the translation types (PmCartesian, PmCylindrical, or PmSpherical), and s indicates a scalar double type.

function

C version C++ version
equality predicate pmXXCompare(X,X) ==, !=
type conversion pmXXConvert(X,X) =
dot product pmXXDot(X,X,X*) dot(X,X)
cross product pmXXCross(X,X,X*) cross(X,X)
magnitude pmXMag(X,s*) mag(X)
normalization (unit) pmXNorm(X,X*) norm(X)
norm predicate pmXIsNorm(X) isNorm(X)
displacement pmXXDisp(X,X,s*) disp(X,X)
sum pmXXAdd(X,X,X*) X+X
difference pmXXSub(X,X,X*) X-X
scalar multiply pmXScalMult(X,s,X*) X*s, s*X
scalar divide pmXScalDiv(X,s,X*) X/s
additive inverse pmXNeg(X,X*) -X
inverse (X dot X-1 = 1) pmXInv(X,X*) inv(X)
projection pmXXProj(X,X,X*) proj(X,X)

 

Rotational Functions

The rotational data representations are used to represent rotations in space of one reference frame relative to another. Functions that operate on these rotational types and combinations of translational and rotational types include comparison, normalization into valid range, checking for normal range, inverse, multiplication by a scalar (scaling the rotation about the constant direction), vector multiply (resulting in a rotation of the vector), and rotational multiply (concatenation of several rotations).

For C functions that take two arguments, the arguments must be the same type. For C++ overloaded functions or operators that take two arguments, they may be of different type. In this case, the constructors are called to do the conversions automatically.

Return values for all the C functions are integer error codes, with the exception of the predicate functions which return 1 if it is true or 0 if not.

Return values for all the C++ functions are the result that corresponds to the last argument of the C functions. The global variable pmErrno should be checked to see if the function was successful.

Following the POSEMATH naming conventions, these functions are prefixed with pm, the abbreviated names of their arguments, and the operation. In the table below, R indicates any of the rotation types (PmRotationVector, PmRotationMatrix, PmQuaternion, PmEulerZyz, PmEulerZyx, or PmRpy), X indicates any of the translation types (PmCartesian, PmCylindrical, or PmSpherical), and s indicates a scalar double type.

function

C version C++ version
equality predicate pmRRCompare(R,R) ==, !=
type conversion pmRRConvert(R,R) =
magnitude pmRMag(R,s*) mag(R)
normalization pmRNorm(R,R*) norm(R)
norm predicate pmRIsNorm(R) isNorm(R)
scalar multiply pmRScalMult(R,s,R*) R*s, s*R
scalar divide pmRScalDiv(R,s,R*) R/s
vector multiply pmRXMult(R,X,X*) R*X, X*R
rotation multiply pmRRMult(R,R,R*) R*R
inverse pmRInv(R,R*) inv(R), -R

 

Combined Functions

The combined representations are used to represent the position and orientation in space of one reference frame relative to another. In POSEMATH, the position and orientation of a reference frame is termed a pose. Although the PmPose type appears to have some preference as a pose representation, both PmPose and PmHomogeneous are pose types. Poses contain both translational and rotational representations. Functions which operate on poses manipulate individual poses, and convert vectors or poses in one reference frame into another. These functions include comparison, normalization into valid range, checking for normal range, inverse, and multiplications between poses and translations, rotations, and other poses.

For C functions which take two arguments, the arguments must be the same type. For C++ overloaded functions or operators which take two arguments, they may be of different type. In this case, the constructors are called to do the conversions automatically.

Return values for all the C functions are integer error codes, with the exception of the predicate functions which return 1 if it is true or 0 if not.

Return values for all the C++ functions are the result that corresponds to the last argument of the C functions. The global variable pmErrno should be checked to see if the function was successful.

Following the POSEMATH naming conventions, these functions are prefixed with pm, the abbreviated names of their arguments, and the operation. In the table below, R indicates any of the translation types (PmCartesian, PmCylindrical, or PmSpherical), X indicates any of the translation types (PmCartesian, PmCylindrical, or PmSpherical), and s indicates a scalar double type.

function

C version C++ version
equality predicate pmPPCompare(P,P) ==, !=
normalization pmPNorm(P,P*) norm(P)
norm predicate pmPIsNorm(R) isNorm(P)
inverse pmPInv(P,P*) inv(P), -P
vector multiply pmPVMult(R,X,X*) R*X, X*R
rotation multiply pmRRMult(R,R,R*) R*R

 

Automatic Conversions in C++

As of this writing, not all of the C conversion functions have been implemented. C programmers should check Appendix B for functions that they intend to use. Some functions whose existence is implied may be stubbed to return PM_IMPL_ERR, or not exist at all. This apparent laziness is due to the combinatorics of the conversion functions, particularly among the rotational types.

For example, there is no C function pmZyzRpyConvert which converts a PmEulerZyz to a PmRpy. The function may not exist, in which case the compiler will flag calls to it, or it may be stubbed to return PM_IMPL_ERR, in which case a run-time error will occur. Ultimately, all the C functions will be implemented in some fashion, but currently they are not. In this case, the C programmer must convert the ZYZ Euler to a rotation matrix, and from a rotation matrix to a roll-pitch-yaw representation, as shown below for the PmEulerZyz zyz and PmRpy rpy:

PmRotationMatrix mat;

pmZyzMatConvert(zyz, &mat);

pmMatRpyConvert(mat, &rpy);

C++ programmers are luckier. Each data representation class has constructors and assignment operators which initialize or assign its representation from any other representation. This may have been accomplished directly, if equivalent C conversion functions exist, or via conversion to some intermediate type. C++ programmers can use the = operator directly:

rpy = zyz;

In this case, the POSEMATH implementation for PM_RPY::operator = (PM_EULER_ZYZ) does the conversion to the intermediate rotation matrix format and out again automatically.

Appendix A: Platform Support and Compilation

The POSEMATH code has been compiled for both Unix and Microsoft platforms, in various specific versions with various compilers. These include:

Operating System Compilers

Sun Solaris Gnu C, C++; CenterLine C++

Linux Gnu C, C++

Real-time Linux Gnu C

LynxOS Gnu C, C++

Microsoft Windows 3.1 Borland C, C++

Microsoft Windows 95 Microsoft Visual C++

Microsoft Windows NT 4.0 Microsoft Visual C++

 

NIST-Specific Compilation

The POSEMATH libraries are available in the RCS platform-specific release directories. The libraries are in the directory

/proj/rcslib/plat/<PLAT>/lib/,

and the header files are located in the directory

/proj/rcslib/plat/<PLAT>/include/. The value for <PLAT> can be one of:

irix5

linux

rtlinux

lynxosPC

lynxosPC23

lynxosVME

sunos4

sunos5

sunos5CC

vxworks5.1

vxworks5.2

Appendix B: Programming Reference

C functions are intended be used by C programmers, since the arguments are of the C types. C++ programmers should use the C++ overloaded operators and functions.

C++ programmers can use the C functions by explicitly converting the C++ types to C types, and back again. toType(src, dst) macros have been provided to accomplish this, where Type is the C abbrevation (e.g., Cart, Cyl, Pose), src is the source to be converted, and dst is the result. These exploit the fact that the fields for both the C and C++ data types have the same names.

Unless indicated otherwise, all functions return an integer error code, which is 0 upon success or one of the error codes defined in posemath.h. This error code is also written to the global variable pmErrno.

There is a bias in the POSEMATH implementation toward using PmCartesian for translation types, PmQuaternion for rotational types, and PmPose for pose types. This is primarily due to the computational efficiency of the quaternion, meaning that programs that use quaternion representations will execute faster than programs using, for example, rotation matrices or Euler angles. In the descriptions of the functions that follow, any translational type can be substituted for PmCartesian, any rotational type can be substituted for PmQuaternion, and any pose type can be substituted for PmPose.

pmQuatCartMult(PmQuaternion q, PmCartesian v, PmCartesian * vout);

Sets vout to the vector resulting from a rotation of v specified by q. Corresponds to C++ operator *, q * v. Operator * is not commutative.

pmQuatQuatMult(PmQuaternion q1, PmQuaternion q2, PmCartesian * qout);

Sets qout to the quaternion representing a rotation first by q2 in the base frame, then by q1 in the q2 frame. Corresponds to C++ operator *, which is not commutative: q1*q2 ¹ q2*q1, in general.

pmPoseCartMult(PmPose p, PmCartesian v, PmCartesian vout);

Sets vout to the vector representing a transformation by the pose p. Corresponds to C++ operator *, p * v. Operator * is not commutative (in fact, v * p yields p whose translation is added by v).

pmPosePoseMult(PmPose p1, PmPose p2, PmPose pout);

Sets pout to the pose representating a transformation first by p1 in the base frame, then by p2 in the p1 frame. Corresponds to C++ operator *, which is not commutative: p1*p2 ¹ p2*p1, in general.

Created July 16, 2014, Updated January 6, 2017