LCOV - code coverage report
Current view: top level - src/AH/Math - Quaternion.hpp (source / functions) Hit Total Coverage
Test: 00f463b534fbea22f0b596e091a60715679e3064 Lines: 115 115 100.0 %
Date: 2024-11-03 16:33:35 Functions: 29 29 100.0 %
Legend: Lines: hit not hit

          Line data    Source code
       1             : /**
       2             :  * @file
       3             :  * @brief    Definition of Quaternion and EulerAngles.
       4             :  * 
       5             :  * Quaternions can be multiplied (Hamiltonian product), normalized and can
       6             :  * perform rotations of vectors. Quaternion also has an implementation of the
       7             :  * following operators:
       8             :  * 
       9             :  *  - `-` (conjugate)
      10             :  *  - `+`, `+=`, `-`, `-=` (Hamiltonian product of quaternions, adds and 
      11             :  *    subtracts angles)
      12             :  *  - `*`, `*=`, `/`, `/=` (multiplication and division by scalars)
      13             :  *  - `==`, `!=` (equality)
      14             :  *  - `<<` (printing)
      15             :  * 
      16             :  * EulerAngles provides the conversions between Euler angles and quaternions.
      17             :  * It also has an implementation of the following operators:
      18             :  * 
      19             :  *  - `==`, `!=` (equality)
      20             :  *  - `<<` (printing)
      21             :  */
      22             : #pragma once
      23             : 
      24             : #include <AH/Arduino-Wrapper.h> // Print
      25             : #include <AH/Math/Degrees.hpp>  // rad2deg()
      26             : #include <AH/Math/Vector.hpp>   // Vec3f
      27             : #include <AH/STL/cmath>         // std::sqrt
      28             : #include <AH/STL/limits>        // std::numeric_limits
      29             : 
      30             : #ifndef ARDUINO
      31             : #include <iosfwd> // std::ostream
      32             : #endif
      33             : 
      34             : BEGIN_AH_NAMESPACE
      35             : 
      36             : /**
      37             :  * @addtogroup  math-types Math Types
      38             :  *              Vector and Quaternion types with the necessary operators and 
      39             :  *              functions.
      40             :  * @{
      41             :  */
      42             : 
      43             : /**
      44             :  * @brief   Type for quaternions of floating point numbers.
      45             :  *
      46             :  * Quaternions can be multiplied (Hamiltonian product), normalized and can
      47             :  * perform rotations of vectors. Quaternion also has an implementation of the
      48             :  * following operators:
      49             :  * 
      50             :  *  - `-` (conjugate)
      51             :  *  - `+`, `+=`, `-`, `-=` (Hamiltonian product of quaternions, adds and 
      52             :  *    subtracts angles)
      53             :  *  - `*`, `*=`, `/`, `/=` (multiplication and division by scalars)
      54             :  *  - `==`, `!=` (equality)
      55             :  *  - `<<` (printing)
      56             :  */
      57             : struct Quaternion {
      58             :     float w = 1.0; ///< Scalar (real) component.
      59             :     float x = 0.0; ///< First vector (imaginary) component @f$ \mathbf{i} @f$.
      60             :     float y = 0.0; ///< Second vector (imaginary) component @f$ \mathbf{j} @f$.
      61             :     float z = 0.0; ///< Third vector (imaginary) component @f$ \mathbf{k} @f$.
      62             : 
      63             :     /// Create a quaternion that is initialized to the identity quaternion.
      64             :     Quaternion() = default;
      65             :     /// Create a quaterion with the given values for w, x, y and z.
      66     1000073 :     Quaternion(float w, float x, float y, float z) : w(w), x(x), y(y), z(z) {}
      67             : 
      68             :     /// Sum of two quaterions uses quaternion multiplication.
      69             :     /// (Composition of the two rotations.)
      70           3 :     Quaternion &operator+=(Quaternion rhs) {
      71           3 :         return *this = hamiltonianProduct(*this, rhs);
      72             :     }
      73             :     /// Sum of two quaternions uses quaternion multiplication.
      74             :     /// (Composition of the two rotations.)
      75           1 :     Quaternion operator+(Quaternion rhs) const {
      76           1 :         return hamiltonianProduct(*this, rhs);
      77             :     }
      78             : 
      79             :     /// Complex conjugate (doesn't change the original quaternion).
      80           4 :     Quaternion conjugated() const { return {w, -x, -y, -z}; }
      81             :     /// Negated quaternion is its conjugate.
      82           3 :     Quaternion operator-() const { return conjugated(); }
      83             : 
      84             :     /// Difference of two quaternions `a` and `b` is the quaternion
      85             :     /// multiplication of `a` and the conjugate of `b`.
      86             :     /// (Composition of the rotation of `a` and the inverse rotation of `b`.)
      87           2 :     Quaternion &operator-=(Quaternion rhs) { return *this += -rhs; }
      88             :     /// Difference of two quaternions `a` and `b` is the quaternion
      89             :     /// multiplication of `a` and the conjugate of `b`.
      90             :     /// (Composition of the rotation of `a` and the inverse rotation of `b`.)
      91           1 :     Quaternion operator-(Quaternion rhs) const {
      92           1 :         Quaternion result = *this;
      93           1 :         result -= rhs;
      94           1 :         return result;
      95             :     }
      96             : 
      97             :     /// Scalar multiplication.
      98           2 :     Quaternion &operator*=(float rhs) {
      99           2 :         w *= rhs;
     100           2 :         x *= rhs;
     101           2 :         y *= rhs;
     102           2 :         z *= rhs;
     103           2 :         return *this;
     104             :     }
     105             :     /// Scalar multiplication.
     106           1 :     Quaternion operator*(float rhs) const {
     107           1 :         Quaternion result = *this;
     108           1 :         result *= rhs;
     109           1 :         return result;
     110             :     }
     111             : 
     112             :     /// Scalar division.
     113     1000016 :     Quaternion &operator/=(float rhs) {
     114     1000016 :         w /= rhs;
     115     1000016 :         x /= rhs;
     116     1000016 :         y /= rhs;
     117     1000016 :         z /= rhs;
     118     1000016 :         return *this;
     119             :     }
     120             :     /// Scalar division.
     121     1000008 :     Quaternion operator/(float rhs) const {
     122     1000008 :         Quaternion result = *this;
     123     1000008 :         result /= rhs;
     124     1000008 :         return result;
     125             :     }
     126             : 
     127             :     /// Norm squared.
     128     1000016 :     float normSquared() const { return w * w + x * x + y * y + z * z; }
     129             :     /// Norm.
     130     1000015 :     float norm() const { return std::sqrt(normSquared()); }
     131             :     /// Normalize this quaternion.
     132           7 :     Quaternion &normalize() { return *this /= norm(); }
     133             :     /// Normalize a copy of this quaternion (doesn't change the original
     134             :     /// quaternion).
     135     1000007 :     Quaternion normalized() const { return *this / norm(); }
     136             : 
     137             :     /**
     138             :      * @brief   Rotate vector by this quaternion.
     139             :      * 
     140             :      * This function uses the normalized version of this quaternion.
     141             :      * 
     142             :      * @note    This function is not the same as `quatrotate` in MATLAB!
     143             :      *          MATLAB rotates by the conjugate of the quaternion, while this
     144             :      *          function rotates by the quaternion itself.
     145             :      */
     146     1000006 :     Vec3f rotate(Vec3f v) const {
     147             :         // Source:
     148             :         // https://en.wikipedia.org/wiki/Quaternions_and_spatial_rotation#Quaternion-derived_rotation_matrix
     149             : 
     150             :         // Compare to the MATLAB convetions:
     151             :         // https://www.mathworks.com/matlabcentral/answers/352465-what-is-the-aerospace-blockset-quaternion-convention
     152             :         // https://www.mathworks.com/help/aerotbx/ug/quatrotate.html#mw_f4c2628b-30a7-4b62-a227-7fb9ef158187
     153             : 
     154             :         // Normalized quaternion.
     155     1000006 :         Quaternion q = normalized();
     156             : 
     157             :         // Rotation matrix.
     158     1000006 :         float M11 = 1 - 2 * q.y * q.y - 2 * q.z * q.z;
     159     1000006 :         float M12 = 2 * (q.x * q.y - q.w * q.z);
     160     1000006 :         float M13 = 2 * (q.x * q.z + q.w * q.y);
     161     1000006 :         float M21 = 2 * (q.x * q.y + q.w * q.z);
     162     1000006 :         float M22 = 1 - 2 * q.x * q.x - 2 * q.z * q.z;
     163     1000006 :         float M23 = 2 * (q.y * q.z - q.w * q.x);
     164     1000006 :         float M31 = 2 * (q.x * q.z - q.w * q.y);
     165     1000006 :         float M32 = 2 * (q.y * q.z + q.w * q.x);
     166     1000006 :         float M33 = 1 - 2 * q.x * q.x - 2 * q.y * q.y;
     167             : 
     168             :         return Vec3f{
     169     1000006 :             M11 * v.x + M12 * v.y + M13 * v.z,
     170     1000006 :             M21 * v.x + M22 * v.y + M23 * v.z,
     171     1000006 :             M31 * v.x + M32 * v.y + M33 * v.z,
     172     3000018 :         };
     173             :     }
     174             : 
     175             :     /// Equality check.
     176          25 :     bool operator==(Quaternion rhs) const {
     177          48 :         return this->w == rhs.w && this->x == rhs.x && this->y == rhs.y &&
     178          48 :                this->z == rhs.z;
     179             :     }
     180             :     /// Inequality check.
     181           2 :     bool operator!=(Quaternion rhs) const { return !(*this == rhs); }
     182             : 
     183             :     /// Identity quaternion (1,0,0,0).
     184           2 :     static Quaternion identity() { return {1, 0, 0, 0}; }
     185             : 
     186             :     /** 
     187             :      * @brief   Calculate the quaternion that satisfies the following: 
     188             :      *          `result.rotate(Vec3f{0, 0, 1}) == v.normalized()`.
     189             :      */
     190     1000005 :     static Quaternion fromDirection(Vec3f v) {
     191             :         /*
     192             :          * Formula:
     193             :          * q = cos(ϑ / 2) + sin(ϑ / 2)·(x·i + y·j + z·k)
     194             :          * where (x y z) is a unit vector representing the axis about which
     195             :          * the body is rotated; ϑ is the angle by which it is rotated.
     196             :          * 
     197             :          * Source: 
     198             :          * https://en.wikipedia.org/wiki/Quaternions_and_spatial_rotation#Using_quaternion_as_rotations
     199             :          * 
     200             :          * The rotational axis (x y z) can be calcuated by taking the normalized
     201             :          * cross product of (0 0 1) and the given vector. The angle of rotation
     202             :          * ϑ can be found using |A×B| = |A||B|·sin(ϑ).
     203             :          */
     204             : 
     205     1000005 :         float eps = std::numeric_limits<float>::epsilon();
     206             : 
     207             :         // Ensure that the norm of v is not zero
     208     1000005 :         float v_norm = v.norm();
     209     1000005 :         if (v_norm <= eps)
     210           1 :             return Quaternion(0, 0, 0, 0); // invalid zero quaternion
     211             : 
     212             :         // Normalize the x and y components (only the sign of v.z matters now)
     213     1000004 :         float x = v.x / v_norm;
     214     1000004 :         float y = v.y / v_norm;
     215             : 
     216             :         // Check the edge case, where v ≃ (0 0 ±1).
     217     1000004 :         if (std::abs(x) <= eps && std::abs(y) <= eps)
     218           1 :             return v.z > 0 ? Quaternion(1, 0, 0, 0) : Quaternion(0, 1, 0, 0);
     219             : 
     220             :         // Calculate the cross product and its norm.
     221             :         // (z component is always zero)
     222     1000003 :         Vec2f cross = {-y, x};
     223     1000003 :         float crossNorm = min(cross.norm(), 1);
     224     1000003 :         cross /= crossNorm;
     225             : 
     226             :         // Calculate the angle ϑ.
     227     1000003 :         float angle = std::asin(crossNorm);
     228     1000003 :         if (v.z < 0)
     229      499712 :             angle = float(180_deg) - angle;
     230             : 
     231             :         // Calculate the resulting quaternion.
     232             :         return {
     233     1000003 :             std::cos(angle / 2),           //
     234     1000003 :             std::sin(angle / 2) * cross.x, //
     235     2000006 :             std::sin(angle / 2) * cross.y, //
     236             :             0,                             // = std::sin(angle / 2) * cross.z
     237     2000006 :         };
     238             :     }
     239             : 
     240             :     /** 
     241             :      * @brief   Calculate the quaternion from a vector that makes a given angle
     242             :      *          with the XZ plane and a given angle with the YZ plane.
     243             :      * 
     244             :      * @param   xAngle
     245             :      *          The angle the vector should make with the XZ plane. A positive
     246             :      *          value represents a positive rotation about the x-axis.
     247             :      * @param   yAngle
     248             :      *          The angle the vector should make with the YZ plane. A positive
     249             :      *          value represents a positive rotation about the y-axis.
     250             :      * 
     251             :      * @return  A quaternion from the vector {tan(yAngle), -tan(xAngle), 1}.
     252             :      */
     253           1 :     static Quaternion fromXYAngle(float xAngle, float yAngle) {
     254           1 :         Vec3f v = {
     255             :             +std::tan(yAngle), // x
     256           1 :             -std::tan(xAngle), // y
     257             :             1,                 // z
     258           2 :         };
     259           3 :         return Quaternion::fromDirection(v);
     260             :     }
     261             : 
     262             :     /// Quaternion multiplication.
     263           5 :     static Quaternion hamiltonianProduct(Quaternion q, Quaternion r) {
     264             :         return {
     265           5 :             r.w * q.w - r.x * q.x - r.y * q.y - r.z * q.z,
     266           5 :             r.w * q.x + r.x * q.w - r.y * q.z + r.z * q.y,
     267           5 :             r.w * q.y + r.x * q.z + r.y * q.w - r.z * q.x,
     268           5 :             r.w * q.z - r.x * q.y + r.y * q.x + r.z * q.w,
     269           5 :         };
     270             :     }
     271             : };
     272             : 
     273             : /// Scalar multiplication.
     274             : /// @related Quaternion
     275           1 : inline Quaternion operator*(float lhs, Quaternion rhs) {
     276           1 :     return {lhs * rhs.w, lhs * rhs.x, lhs * rhs.y, lhs * rhs.z};
     277             : }
     278             : 
     279             : /**
     280             :  * @brief  Struct for Euler angles of floating point numbers.
     281             :  * 
     282             :  * EulerAngles provides the conversions between Euler angles and quaternions.
     283             :  * It also has an implementation of the following operators:
     284             :  * 
     285             :  *  - `==`, `!=` (equality)
     286             :  *  - `<<` (printing)
     287             :  */
     288             : struct EulerAngles {
     289             :     float yaw = 0.0;   ///< Z : drone Z = world +Z.
     290             :     float pitch = 0.0; ///< Y': drone Y = world -X.
     291             :     float roll = 0.0;  ///< X": drone X = world +Y.
     292             : 
     293             :     /// Create Euler angles that are initialized to (0 0 0), or upright.
     294             :     EulerAngles() = default;
     295             :     /// Create Euler angles with the given values for yaw, pitch and roll.
     296          11 :     EulerAngles(float yaw, float pitch, float roll)
     297          11 :         : yaw(yaw), pitch(pitch), roll(roll) {}
     298             :     /// Create Euler angles from the given quaternion.
     299           3 :     EulerAngles(Quaternion q) : EulerAngles{quat2eul(q)} {}
     300             : 
     301             :     /// Implicitly convert these Euler angles to a quaternion.
     302           3 :     operator Quaternion() const { return eul2quat(*this); }
     303             : 
     304             :     /// Equality check.
     305           1 :     bool operator==(EulerAngles rhs) const {
     306           2 :         return this->yaw == rhs.yaw && this->pitch == rhs.pitch &&
     307           2 :                this->roll == rhs.roll;
     308             :     }
     309             :     /// Inequality check.
     310             :     bool operator!=(EulerAngles rhs) const { return !(*this == rhs); }
     311             : 
     312             :     /// Convert the given quaternion to Euler angles.
     313           5 :     static EulerAngles quat2eul(Quaternion q) {
     314             :         // Source: https://en.wikipedia.org/wiki/Conversion_between_quaternions_and_Euler_angles#Quaternion_to_Euler_Angles_Conversion
     315             : 
     316             :         // Normalize quaternion (q is passed by value, so this is a copy of q).
     317           5 :         q.normalize();
     318             : 
     319           5 :         float phi = std::atan2(2 * (q.w * q.x + q.y * q.z),
     320           5 :                                1 - 2 * (q.x * q.x + q.y * q.y));
     321           5 :         float theta = std::asin(2 * (q.w * q.y - q.z * q.x));
     322           5 :         float psi = std::atan2(2 * (q.w * q.z + q.x * q.y),
     323           5 :                                1 - 2 * (q.y * q.y + q.z * q.z));
     324           5 :         return {psi, theta, phi};
     325             :     }
     326             : 
     327             :     /// Convert the given Euler angles to a quaternion.
     328           5 :     static Quaternion eul2quat(EulerAngles eulerAngles) {
     329             :         // Source: https://en.wikipedia.org/wiki/Conversion_between_quaternions_and_Euler_angles#Euler_Angles_to_Quaternion_Conversion
     330             : 
     331           5 :         float cy = std::cos(eulerAngles.yaw / 2);
     332           5 :         float sy = std::sin(eulerAngles.yaw / 2);
     333           5 :         float cp = std::cos(eulerAngles.pitch / 2);
     334           5 :         float sp = std::sin(eulerAngles.pitch / 2);
     335           5 :         float cr = std::cos(eulerAngles.roll / 2);
     336           5 :         float sr = std::sin(eulerAngles.roll / 2);
     337             : 
     338             :         return {
     339           5 :             cy * cp * cr + sy * sp * sr,
     340           5 :             cy * cp * sr - sy * sp * cr,
     341           5 :             sy * cp * sr + cy * sp * cr,
     342           5 :             sy * cp * cr - cy * sp * sr,
     343           5 :         };
     344             :     }
     345             : };
     346             : 
     347             : #ifndef ARDUINO
     348             : 
     349             : /// Printing.
     350             : /// @related  Quaternion
     351             : std::ostream &operator<<(std::ostream &os, Quaternion q);
     352             : 
     353             : /// Printing.
     354             : /// @related  EulerAngles
     355             : std::ostream &operator<<(std::ostream &os, EulerAngles e);
     356             : 
     357             : #endif // ARDUINO
     358             : 
     359             : /// Printing.
     360             : /// @related  Quaternion
     361             : Print &operator<<(Print &os, Quaternion e);
     362             : 
     363             : /// Printing.
     364             : /// @related  EulerAngles
     365             : Print &operator<<(Print &os, EulerAngles e);
     366             : 
     367             : /// @}
     368             : 
     369             : END_AH_NAMESPACE

Generated by: LCOV version 1.15