Refactor rotations. Add Rotation matrix to Euler conversion
This commit is contained in:
parent
67994d35d9
commit
eff70eb690
@ -7,7 +7,7 @@ package space.kscience.kmath.geometry
|
|||||||
|
|
||||||
import space.kscience.kmath.linear.Point
|
import space.kscience.kmath.linear.Point
|
||||||
|
|
||||||
public interface Vector2D<T> : Point<T>, Vector {
|
public interface Vector2D<T> : Point<T> {
|
||||||
public val x: T
|
public val x: T
|
||||||
public val y: T
|
public val y: T
|
||||||
override val size: Int get() = 2
|
override val size: Int get() = 2
|
||||||
|
@ -8,7 +8,7 @@ package space.kscience.kmath.geometry
|
|||||||
import space.kscience.kmath.linear.Point
|
import space.kscience.kmath.linear.Point
|
||||||
import space.kscience.kmath.structures.Buffer
|
import space.kscience.kmath.structures.Buffer
|
||||||
|
|
||||||
public interface Vector3D<T> : Point<T>, Vector {
|
public interface Vector3D<T> : Point<T> {
|
||||||
public val x: T
|
public val x: T
|
||||||
public val y: T
|
public val y: T
|
||||||
public val z: T
|
public val z: T
|
||||||
|
@ -6,8 +6,6 @@
|
|||||||
package space.kscience.kmath.geometry.euclidean2d
|
package space.kscience.kmath.geometry.euclidean2d
|
||||||
|
|
||||||
import kotlinx.serialization.Serializable
|
import kotlinx.serialization.Serializable
|
||||||
import space.kscience.kmath.geometry.Euclidean2DSpace.distanceTo
|
|
||||||
import kotlin.math.*
|
|
||||||
import kotlin.math.PI
|
import kotlin.math.PI
|
||||||
|
|
||||||
/**
|
/**
|
||||||
@ -16,7 +14,7 @@ import kotlin.math.PI
|
|||||||
@Serializable
|
@Serializable
|
||||||
public data class Circle2D(
|
public data class Circle2D(
|
||||||
@Serializable(Float64Space2D.VectorSerializer::class) public val center: DoubleVector2D,
|
@Serializable(Float64Space2D.VectorSerializer::class) public val center: DoubleVector2D,
|
||||||
public val radius: Double
|
public val radius: Double,
|
||||||
)
|
)
|
||||||
|
|
||||||
|
|
||||||
|
@ -14,17 +14,15 @@ import kotlinx.serialization.encoding.Encoder
|
|||||||
import space.kscience.kmath.geometry.GeometrySpace
|
import space.kscience.kmath.geometry.GeometrySpace
|
||||||
import space.kscience.kmath.geometry.Vector2D
|
import space.kscience.kmath.geometry.Vector2D
|
||||||
import space.kscience.kmath.operations.Float32Field
|
import space.kscience.kmath.operations.Float32Field
|
||||||
import space.kscience.kmath.operations.ScaleOperations
|
import space.kscience.kmath.structures.Float32
|
||||||
import kotlin.math.pow
|
import kotlin.math.pow
|
||||||
import kotlin.math.sqrt
|
import kotlin.math.sqrt
|
||||||
|
|
||||||
@Serializable(Float32Space2D.VectorSerializer::class)
|
@Serializable(Float32Space2D.VectorSerializer::class)
|
||||||
public interface Float32Vector2D: Vector2D<Float>
|
public interface Float32Vector2D : Vector2D<Float>
|
||||||
|
|
||||||
|
|
||||||
public object Float32Space2D :
|
public object Float32Space2D : GeometrySpace<Float32Vector2D, Float32> {
|
||||||
GeometrySpace<Float32Vector2D>,
|
|
||||||
ScaleOperations<Float32Vector2D> {
|
|
||||||
|
|
||||||
@Serializable
|
@Serializable
|
||||||
@SerialName("Float32Vector2D")
|
@SerialName("Float32Vector2D")
|
||||||
@ -53,13 +51,13 @@ public object Float32Space2D :
|
|||||||
|
|
||||||
override val zero: Float32Vector2D by lazy { vector(0f, 0f) }
|
override val zero: Float32Vector2D by lazy { vector(0f, 0f) }
|
||||||
|
|
||||||
override fun norm(arg: Float32Vector2D): Double = sqrt(arg.x.pow(2) + arg.y.pow(2)).toDouble()
|
override fun norm(arg: Float32Vector2D): Float32 = sqrt(arg.x.pow(2) + arg.y.pow(2))
|
||||||
|
|
||||||
public fun Float32Vector2D.norm(): Double = norm(this)
|
public fun Float32Vector2D.norm(): Float32 = norm(this)
|
||||||
|
|
||||||
override fun Float32Vector2D.unaryMinus(): Float32Vector2D = vector(-x, -y)
|
override fun Float32Vector2D.unaryMinus(): Float32Vector2D = vector(-x, -y)
|
||||||
|
|
||||||
override fun Float32Vector2D.distanceTo(other: Float32Vector2D): Double = (this - other).norm()
|
override fun Float32Vector2D.distanceTo(other: Float32Vector2D): Float32 = (this - other).norm()
|
||||||
|
|
||||||
override fun add(left: Float32Vector2D, right: Float32Vector2D): Float32Vector2D =
|
override fun add(left: Float32Vector2D, right: Float32Vector2D): Float32Vector2D =
|
||||||
vector(left.x + right.x, left.y + right.y)
|
vector(left.x + right.x, left.y + right.y)
|
||||||
@ -72,6 +70,8 @@ public object Float32Space2D :
|
|||||||
|
|
||||||
public val xAxis: Float32Vector2D = vector(1.0, 0.0)
|
public val xAxis: Float32Vector2D = vector(1.0, 0.0)
|
||||||
public val yAxis: Float32Vector2D = vector(0.0, 1.0)
|
public val yAxis: Float32Vector2D = vector(0.0, 1.0)
|
||||||
|
|
||||||
|
override val defaultPrecision: Float32 = 1e-3f
|
||||||
}
|
}
|
||||||
|
|
||||||
public fun Float32Vector2D(x: Number, y: Number): Float32Vector2D = Float32Space2D.vector(x, y)
|
public fun Float32Vector2D(x: Number, y: Number): Float32Vector2D = Float32Space2D.vector(x, y)
|
||||||
|
@ -13,14 +13,12 @@ import kotlinx.serialization.encoding.Decoder
|
|||||||
import kotlinx.serialization.encoding.Encoder
|
import kotlinx.serialization.encoding.Encoder
|
||||||
import space.kscience.kmath.geometry.GeometrySpace
|
import space.kscience.kmath.geometry.GeometrySpace
|
||||||
import space.kscience.kmath.geometry.Vector2D
|
import space.kscience.kmath.geometry.Vector2D
|
||||||
|
import space.kscience.kmath.linear.Float64LinearSpace
|
||||||
import space.kscience.kmath.operations.Float64Field
|
import space.kscience.kmath.operations.Float64Field
|
||||||
import space.kscience.kmath.operations.Norm
|
|
||||||
import space.kscience.kmath.operations.ScaleOperations
|
|
||||||
import kotlin.math.pow
|
import kotlin.math.pow
|
||||||
import kotlin.math.sqrt
|
import kotlin.math.sqrt
|
||||||
|
|
||||||
|
|
||||||
|
|
||||||
public typealias DoubleVector2D = Vector2D<Double>
|
public typealias DoubleVector2D = Vector2D<Double>
|
||||||
public typealias Float64Vector2D = Vector2D<Double>
|
public typealias Float64Vector2D = Vector2D<Double>
|
||||||
|
|
||||||
@ -30,10 +28,9 @@ public val Vector2D<Double>.r: Double get() = Float64Space2D.norm(this)
|
|||||||
/**
|
/**
|
||||||
* 2D Euclidean space
|
* 2D Euclidean space
|
||||||
*/
|
*/
|
||||||
public object Float64Space2D :
|
public object Float64Space2D : GeometrySpace<DoubleVector2D, Double> {
|
||||||
GeometrySpace<DoubleVector2D, Double>,
|
|
||||||
ScaleOperations<DoubleVector2D>,
|
public val linearSpace: Float64LinearSpace = Float64LinearSpace
|
||||||
Norm<DoubleVector2D, Double> {
|
|
||||||
|
|
||||||
@Serializable
|
@Serializable
|
||||||
@SerialName("Float64Vector2D")
|
@SerialName("Float64Vector2D")
|
||||||
|
@ -14,17 +14,15 @@ import kotlinx.serialization.encoding.Encoder
|
|||||||
import space.kscience.kmath.geometry.GeometrySpace
|
import space.kscience.kmath.geometry.GeometrySpace
|
||||||
import space.kscience.kmath.geometry.Vector3D
|
import space.kscience.kmath.geometry.Vector3D
|
||||||
import space.kscience.kmath.operations.Float32Field
|
import space.kscience.kmath.operations.Float32Field
|
||||||
import space.kscience.kmath.operations.ScaleOperations
|
import space.kscience.kmath.structures.Float32
|
||||||
import kotlin.math.pow
|
import kotlin.math.pow
|
||||||
import kotlin.math.sqrt
|
import kotlin.math.sqrt
|
||||||
|
|
||||||
@Serializable(Float32Space3D.VectorSerializer::class)
|
@Serializable(Float32Space3D.VectorSerializer::class)
|
||||||
public interface Float32Vector3D: Vector3D<Float>
|
public interface Float32Vector3D : Vector3D<Float>
|
||||||
|
|
||||||
|
|
||||||
public object Float32Space3D :
|
public object Float32Space3D : GeometrySpace<Float32Vector3D, Float32> {
|
||||||
GeometrySpace<Float32Vector3D>,
|
|
||||||
ScaleOperations<Float32Vector3D>{
|
|
||||||
|
|
||||||
@Serializable
|
@Serializable
|
||||||
@SerialName("Float32Vector3D")
|
@SerialName("Float32Vector3D")
|
||||||
@ -54,13 +52,13 @@ public object Float32Space3D :
|
|||||||
|
|
||||||
override val zero: Float32Vector3D by lazy { vector(0.0, 0.0, 0.0) }
|
override val zero: Float32Vector3D by lazy { vector(0.0, 0.0, 0.0) }
|
||||||
|
|
||||||
override fun norm(arg: Float32Vector3D): Double = sqrt(arg.x.pow(2) + arg.y.pow(2) + arg.z.pow(2)).toDouble()
|
override fun norm(arg: Float32Vector3D): Float32 = sqrt(arg.x.pow(2) + arg.y.pow(2) + arg.z.pow(2))
|
||||||
|
|
||||||
public fun Float32Vector3D.norm(): Double = norm(this)
|
public fun Float32Vector3D.norm(): Float32 = norm(this)
|
||||||
|
|
||||||
override fun Float32Vector3D.unaryMinus(): Float32Vector3D = vector(-x, -y, -z)
|
override fun Float32Vector3D.unaryMinus(): Float32Vector3D = vector(-x, -y, -z)
|
||||||
|
|
||||||
override fun Float32Vector3D.distanceTo(other: Float32Vector3D): Double = (this - other).norm()
|
override fun Float32Vector3D.distanceTo(other: Float32Vector3D): Float32 = (this - other).norm()
|
||||||
|
|
||||||
override fun add(left: Float32Vector3D, right: Float32Vector3D): Float32Vector3D =
|
override fun add(left: Float32Vector3D, right: Float32Vector3D): Float32Vector3D =
|
||||||
vector(left.x + right.x, left.y + right.y, left.z + right.z)
|
vector(left.x + right.x, left.y + right.y, left.z + right.z)
|
||||||
@ -101,6 +99,8 @@ public object Float32Space3D :
|
|||||||
public val xAxis: Float32Vector3D = vector(1.0, 0.0, 0.0)
|
public val xAxis: Float32Vector3D = vector(1.0, 0.0, 0.0)
|
||||||
public val yAxis: Float32Vector3D = vector(0.0, 1.0, 0.0)
|
public val yAxis: Float32Vector3D = vector(0.0, 1.0, 0.0)
|
||||||
public val zAxis: Float32Vector3D = vector(0.0, 0.0, 1.0)
|
public val zAxis: Float32Vector3D = vector(0.0, 0.0, 1.0)
|
||||||
|
|
||||||
|
override val defaultPrecision: Float32 = 1e-3f
|
||||||
}
|
}
|
||||||
|
|
||||||
public fun Float32Vector3D(x: Number, y: Number, z: Number): Float32Vector3D = Float32Space3D.vector(x, y, z)
|
public fun Float32Vector3D(x: Number, y: Number, z: Number): Float32Vector3D = Float32Space3D.vector(x, y, z)
|
||||||
|
@ -13,9 +13,8 @@ import kotlinx.serialization.encoding.Decoder
|
|||||||
import kotlinx.serialization.encoding.Encoder
|
import kotlinx.serialization.encoding.Encoder
|
||||||
import space.kscience.kmath.geometry.GeometrySpace
|
import space.kscience.kmath.geometry.GeometrySpace
|
||||||
import space.kscience.kmath.geometry.Vector3D
|
import space.kscience.kmath.geometry.Vector3D
|
||||||
|
import space.kscience.kmath.linear.Float64LinearSpace
|
||||||
import space.kscience.kmath.operations.Float64Field
|
import space.kscience.kmath.operations.Float64Field
|
||||||
import space.kscience.kmath.operations.Norm
|
|
||||||
import space.kscience.kmath.operations.ScaleOperations
|
|
||||||
import kotlin.math.pow
|
import kotlin.math.pow
|
||||||
import kotlin.math.sqrt
|
import kotlin.math.sqrt
|
||||||
|
|
||||||
@ -37,8 +36,9 @@ public typealias Float64Vector3D = Vector3D<Double>
|
|||||||
|
|
||||||
public val DoubleVector3D.r: Double get() = Float64Space3D.norm(this)
|
public val DoubleVector3D.r: Double get() = Float64Space3D.norm(this)
|
||||||
|
|
||||||
public object Float64Space3D : GeometrySpace<DoubleVector3D, Double>, ScaleOperations<DoubleVector3D>,
|
public object Float64Space3D : GeometrySpace<DoubleVector3D, Double>{
|
||||||
Norm<DoubleVector3D, Double> {
|
|
||||||
|
public val linearSpace: Float64LinearSpace = Float64LinearSpace
|
||||||
|
|
||||||
@Serializable
|
@Serializable
|
||||||
@SerialName("Float64Vector3D")
|
@SerialName("Float64Vector3D")
|
||||||
|
@ -6,20 +6,36 @@
|
|||||||
package space.kscience.kmath.geometry.euclidean3d
|
package space.kscience.kmath.geometry.euclidean3d
|
||||||
|
|
||||||
import space.kscience.kmath.UnstableKMathAPI
|
import space.kscience.kmath.UnstableKMathAPI
|
||||||
import space.kscience.kmath.complex.Quaternion
|
import space.kscience.kmath.complex.*
|
||||||
import space.kscience.kmath.complex.QuaternionAlgebra
|
|
||||||
import space.kscience.kmath.complex.normalized
|
|
||||||
import space.kscience.kmath.complex.reciprocal
|
|
||||||
import space.kscience.kmath.geometry.*
|
import space.kscience.kmath.geometry.*
|
||||||
import space.kscience.kmath.linear.LinearSpace
|
import space.kscience.kmath.linear.LinearSpace
|
||||||
import space.kscience.kmath.linear.Matrix
|
import space.kscience.kmath.linear.Matrix
|
||||||
import space.kscience.kmath.linear.linearSpace
|
import space.kscience.kmath.linear.linearSpace
|
||||||
import space.kscience.kmath.linear.matrix
|
import space.kscience.kmath.linear.matrix
|
||||||
import space.kscience.kmath.operations.Float64Field
|
import space.kscience.kmath.operations.Float64Field
|
||||||
import kotlin.math.pow
|
import kotlin.math.*
|
||||||
import kotlin.math.sqrt
|
|
||||||
|
|
||||||
internal fun DoubleVector3D.toQuaternion(): Quaternion = Quaternion(0.0, x, y, z)
|
public operator fun Quaternion.times(other: Quaternion): Quaternion = QuaternionAlgebra.multiply(this, other)
|
||||||
|
|
||||||
|
public operator fun Quaternion.div(other: Quaternion): Quaternion = QuaternionAlgebra.divide(this, other)
|
||||||
|
|
||||||
|
public fun Quaternion.power(number: Number): Quaternion = QuaternionAlgebra.power(this, number)
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Linear interpolation between [from] and [to] in spherical space
|
||||||
|
*/
|
||||||
|
public fun QuaternionAlgebra.slerp(from: Quaternion, to: Quaternion, fraction: Double): Quaternion =
|
||||||
|
(to / from).pow(fraction) * from
|
||||||
|
|
||||||
|
public fun QuaternionAlgebra.angleBetween(q1: Quaternion, q2: Quaternion): Angle = (q1.conjugate * q2).theta
|
||||||
|
|
||||||
|
public infix fun Quaternion.dot(other: Quaternion): Double = w * other.w + x * other.x + y * other.y + z * other.z
|
||||||
|
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Represent a vector as quaternion with zero a rotation angle.
|
||||||
|
*/
|
||||||
|
internal fun DoubleVector3D.asQuaternion(): Quaternion = Quaternion(0.0, x, y, z)
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* Angle in radians denoted by this quaternion rotation
|
* Angle in radians denoted by this quaternion rotation
|
||||||
@ -51,23 +67,30 @@ public val Quaternion.vector: DoubleVector3D
|
|||||||
}
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* Rotate a vector in a [Float64Space3D]
|
* Rotate a vector in a [Float64Space3D] with [quaternion]
|
||||||
*/
|
*/
|
||||||
public fun Float64Space3D.rotate(vector: DoubleVector3D, q: Quaternion): DoubleVector3D = with(QuaternionAlgebra) {
|
public fun Float64Space3D.rotate(vector: DoubleVector3D, quaternion: Quaternion): DoubleVector3D =
|
||||||
val p = vector.toQuaternion()
|
with(QuaternionAlgebra) {
|
||||||
(q * p * q.reciprocal).vector
|
val p = vector.asQuaternion()
|
||||||
}
|
(quaternion * p * quaternion.reciprocal).vector
|
||||||
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* Use a composition of quaternions to create a rotation
|
* Use a composition of quaternions to create a rotation
|
||||||
*/
|
*/
|
||||||
@UnstableKMathAPI
|
@UnstableKMathAPI
|
||||||
public fun Float64Space3D.rotate(vector: DoubleVector3D, composition: QuaternionAlgebra.() -> Quaternion): DoubleVector3D =
|
public fun Float64Space3D.rotate(
|
||||||
|
vector: DoubleVector3D,
|
||||||
|
composition: QuaternionAlgebra.() -> Quaternion,
|
||||||
|
): DoubleVector3D =
|
||||||
rotate(vector, QuaternionAlgebra.composition())
|
rotate(vector, QuaternionAlgebra.composition())
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Rotate a [Float64] vector in 3D space with a rotation matrix
|
||||||
|
*/
|
||||||
public fun Float64Space3D.rotate(vector: DoubleVector3D, matrix: Matrix<Double>): DoubleVector3D {
|
public fun Float64Space3D.rotate(vector: DoubleVector3D, matrix: Matrix<Double>): DoubleVector3D {
|
||||||
require(matrix.colNum == 3 && matrix.rowNum == 3) { "Square 3x3 rotation matrix is required" }
|
require(matrix.colNum == 3 && matrix.rowNum == 3) { "Square 3x3 rotation matrix is required" }
|
||||||
return with(Float64Field.linearSpace) { matrix.dot(vector).asVector3D() }
|
return with(linearSpace) { (matrix dot vector).asVector3D() }
|
||||||
}
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
@ -86,6 +109,8 @@ public fun Quaternion.toRotationMatrix(
|
|||||||
}
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
|
* Convert a quaternion to a rotation matrix
|
||||||
|
*
|
||||||
* taken from https://www.euclideanspace.com/maths/geometry/rotations/conversions/matrixToQuaternion/
|
* taken from https://www.euclideanspace.com/maths/geometry/rotations/conversions/matrixToQuaternion/
|
||||||
*/
|
*/
|
||||||
public fun Quaternion.Companion.fromRotationMatrix(matrix: Matrix<Double>): Quaternion {
|
public fun Quaternion.Companion.fromRotationMatrix(matrix: Matrix<Double>): Quaternion {
|
||||||
@ -146,6 +171,8 @@ public enum class RotationOrder {
|
|||||||
}
|
}
|
||||||
|
|
||||||
/**
|
/**
|
||||||
|
* Create a quaternion from Euler angles
|
||||||
|
*
|
||||||
* Based on https://github.com/mrdoob/three.js/blob/master/src/math/Quaternion.js
|
* Based on https://github.com/mrdoob/three.js/blob/master/src/math/Quaternion.js
|
||||||
*/
|
*/
|
||||||
public fun Quaternion.Companion.fromEuler(
|
public fun Quaternion.Companion.fromEuler(
|
||||||
@ -154,13 +181,13 @@ public fun Quaternion.Companion.fromEuler(
|
|||||||
c: Angle,
|
c: Angle,
|
||||||
rotationOrder: RotationOrder,
|
rotationOrder: RotationOrder,
|
||||||
): Quaternion {
|
): Quaternion {
|
||||||
val c1 = cos (a / 2)
|
val c1 = cos(a / 2)
|
||||||
val c2 = cos (b / 2)
|
val c2 = cos(b / 2)
|
||||||
val c3 = cos (c / 2)
|
val c3 = cos(c / 2)
|
||||||
|
|
||||||
val s1 = sin (a / 2)
|
val s1 = sin(a / 2)
|
||||||
val s2 = sin (b / 2)
|
val s2 = sin(b / 2)
|
||||||
val s3 = sin (c / 2)
|
val s3 = sin(c / 2)
|
||||||
|
|
||||||
return when (rotationOrder) {
|
return when (rotationOrder) {
|
||||||
|
|
||||||
@ -206,6 +233,130 @@ public fun Quaternion.Companion.fromEuler(
|
|||||||
c1 * s2 * c3 - s1 * c2 * s3,
|
c1 * s2 * c3 - s1 * c2 * s3,
|
||||||
c1 * c2 * s3 + s1 * s2 * c3
|
c1 * c2 * s3 + s1 * s2 * c3
|
||||||
)
|
)
|
||||||
else -> TODO("Proper Euler rotation orders are not supported yet")
|
|
||||||
|
else -> TODO("Proper Euler rotation orders are not supported yet")
|
||||||
}
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
/**
|
||||||
|
* A vector consisting of angles
|
||||||
|
*/
|
||||||
|
public data class AngleVector(override val x: Angle, override val y: Angle, override val z: Angle) : Vector3D<Angle> {
|
||||||
|
public companion object
|
||||||
|
}
|
||||||
|
|
||||||
|
public fun Quaternion.Companion.fromEuler(
|
||||||
|
angles: AngleVector,
|
||||||
|
rotationOrder: RotationOrder,
|
||||||
|
): Quaternion = fromEuler(angles.x, angles.y, angles.z, rotationOrder)
|
||||||
|
|
||||||
|
/**
|
||||||
|
* Based on https://github.com/mrdoob/three.js/blob/master/src/math/Euler.js
|
||||||
|
*/
|
||||||
|
public fun AngleVector.Companion.fromRotationMatrix(
|
||||||
|
matrix: Matrix<Double>,
|
||||||
|
rotationOrder: RotationOrder,
|
||||||
|
gimbaldLockThreshold: Double = 0.9999999,
|
||||||
|
): AngleVector = when (rotationOrder) {
|
||||||
|
|
||||||
|
RotationOrder.XYZ -> {
|
||||||
|
if (abs(matrix[0, 2]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
atan2(-matrix[1, 2], matrix[2, 2]).radians,
|
||||||
|
asin(matrix[0, 2].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
atan2(-matrix[0, 1], matrix[0, 0]).radians
|
||||||
|
)
|
||||||
|
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
atan2(matrix[2, 1], matrix[1, 1]).radians,
|
||||||
|
asin(matrix[0, 2].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
Angle.zero
|
||||||
|
)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
RotationOrder.YXZ -> {
|
||||||
|
if (abs(matrix[1, 2]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
x = asin(-matrix[1, 2].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
y = atan2(matrix[0, 2], matrix[2, 2]).radians,
|
||||||
|
z = atan2(matrix[1, 0], matrix[1, 1]).radians,
|
||||||
|
)
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
x = asin(-matrix[1, 2].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
y = atan2(-matrix[2, 0], matrix[0, 0]).radians,
|
||||||
|
z = Angle.zero,
|
||||||
|
)
|
||||||
|
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
RotationOrder.ZXY -> {
|
||||||
|
if (abs(matrix[2, 1]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
x = asin(matrix[2, 1].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
y = atan2(-matrix[2, 0], matrix[2, 2]).radians,
|
||||||
|
z = atan2(-matrix[0, 1], matrix[1, 1]).radians,
|
||||||
|
)
|
||||||
|
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
x = asin(matrix[2, 1].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
y = Angle.zero,
|
||||||
|
z = atan2(matrix[1, 0], matrix[0, 0]).radians,
|
||||||
|
)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
RotationOrder.ZYX -> {
|
||||||
|
if (abs(matrix[2, 0]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
x = atan2(matrix[2, 1], matrix[2, 2]).radians,
|
||||||
|
y = asin(-matrix[2, 0].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
z = atan2(matrix[1, 0], matrix[0, 0]).radians,
|
||||||
|
)
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
x = Angle.zero,
|
||||||
|
y = asin(-matrix[2, 0].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
z = atan2(-matrix[0, 1], matrix[1, 1]).radians,
|
||||||
|
)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
RotationOrder.YZX -> {
|
||||||
|
if (abs(matrix[1, 0]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
x = atan2(-matrix[1, 2], matrix[1, 1]).radians,
|
||||||
|
y = atan2(-matrix[2, 0], matrix[0, 0]).radians,
|
||||||
|
z = asin(matrix[1, 0].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
)
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
x = Angle.zero,
|
||||||
|
y = atan2(matrix[0, 2], matrix[2, 2]).radians,
|
||||||
|
z = asin(matrix[1, 0].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
RotationOrder.XZY -> {
|
||||||
|
if (abs(matrix[0, 1]) < gimbaldLockThreshold) {
|
||||||
|
AngleVector(
|
||||||
|
x = atan2(matrix[2, 1], matrix[1, 1]).radians,
|
||||||
|
y = atan2(matrix[0, 2], matrix[0, 0]).radians,
|
||||||
|
z = asin(-matrix[0, 1].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
)
|
||||||
|
} else {
|
||||||
|
AngleVector(
|
||||||
|
x = atan2(-matrix[1, 2], matrix[2, 2]).radians,
|
||||||
|
y = Angle.zero,
|
||||||
|
z = asin(-matrix[0, 1].coerceIn(-1.0, 1.0)).radians,
|
||||||
|
)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
|
else -> TODO("Proper Euler rotation orders are not supported yet")
|
||||||
}
|
}
|
@ -1,55 +0,0 @@
|
|||||||
/*
|
|
||||||
* Copyright 2018-2023 KMath contributors.
|
|
||||||
* Use of this source code is governed by the Apache 2.0 license that can be found in the license/LICENSE.txt file.
|
|
||||||
*/
|
|
||||||
|
|
||||||
package space.kscience.kmath.geometry
|
|
||||||
|
|
||||||
import space.kscience.kmath.complex.Quaternion
|
|
||||||
import space.kscience.kmath.complex.QuaternionAlgebra
|
|
||||||
import space.kscience.kmath.complex.conjugate
|
|
||||||
import space.kscience.kmath.complex.normalized
|
|
||||||
import space.kscience.kmath.geometry.euclidean3d.Float32Space3D
|
|
||||||
import space.kscience.kmath.geometry.euclidean3d.Float32Vector3D
|
|
||||||
import space.kscience.kmath.geometry.euclidean3d.theta
|
|
||||||
import kotlin.math.asin
|
|
||||||
import kotlin.math.atan2
|
|
||||||
import kotlin.math.pow
|
|
||||||
|
|
||||||
public operator fun Quaternion.times(other: Quaternion): Quaternion = QuaternionAlgebra.multiply(this, other)
|
|
||||||
|
|
||||||
public operator fun Quaternion.div(other: Quaternion): Quaternion = QuaternionAlgebra.divide(this, other)
|
|
||||||
|
|
||||||
public fun Quaternion.power(number: Number): Quaternion = QuaternionAlgebra.power(this, number)
|
|
||||||
|
|
||||||
/**
|
|
||||||
* Linear interpolation between [from] and [to] in spherical space
|
|
||||||
*/
|
|
||||||
public fun QuaternionAlgebra.slerp(from: Quaternion, to: Quaternion, fraction: Double): Quaternion =
|
|
||||||
(to / from).pow(fraction) * from
|
|
||||||
|
|
||||||
public fun QuaternionAlgebra.angleBetween(q1: Quaternion, q2: Quaternion): Angle = (q1.conjugate * q2).theta
|
|
||||||
|
|
||||||
public val Quaternion.inclination: Radians get() = asin(2 * (w * y - x * z)).radians
|
|
||||||
|
|
||||||
public val Quaternion.azimuth: Angle get() = atan2(2 * (w * z + x * y), 1 - 2 * (y.pow(2) + z.pow(2))).radians.normalized()
|
|
||||||
|
|
||||||
public infix fun Quaternion.dot(other: Quaternion): Double = w * other.w + x * other.x + y * other.y + z * other.z
|
|
||||||
|
|
||||||
|
|
||||||
private fun Quaternion.normalizedToEuler(): Float32Vector3D {
|
|
||||||
val roll = atan2(2 * y * w + 2 * x * z, 1 - 2 * y * y - 2 * z * z);
|
|
||||||
val pitch = atan2(2 * x * w - 2 * y * z, 1 - 2 * x * x - 2 * z * z);
|
|
||||||
val yaw = asin(2 * x * y + 2 * z * w);
|
|
||||||
|
|
||||||
return Float32Vector3D(roll, pitch, yaw)
|
|
||||||
}
|
|
||||||
|
|
||||||
/**
|
|
||||||
* Quaternion to XYZ Cardan angles
|
|
||||||
*/
|
|
||||||
public fun Quaternion.toEuler(): Float32Vector3D = if (QuaternionAlgebra.norm(this) == 0.0) {
|
|
||||||
Float32Space3D.zero
|
|
||||||
} else {
|
|
||||||
normalized().normalizedToEuler()
|
|
||||||
}
|
|
@ -5,6 +5,11 @@
|
|||||||
|
|
||||||
package space.kscience.kmath.geometry
|
package space.kscience.kmath.geometry
|
||||||
|
|
||||||
|
import space.kscience.kmath.geometry.euclidean2d.Float64Space2D
|
||||||
|
import space.kscience.kmath.geometry.euclidean2d.Float64Vector2D
|
||||||
|
import space.kscience.kmath.geometry.euclidean3d.Float64Space3D
|
||||||
|
import space.kscience.kmath.geometry.euclidean3d.Float64Vector3D
|
||||||
|
|
||||||
|
|
||||||
/**
|
/**
|
||||||
* Vector equality within given [precision] (using [GeometrySpace.norm] provided by the space
|
* Vector equality within given [precision] (using [GeometrySpace.norm] provided by the space
|
||||||
@ -22,7 +27,7 @@ public fun <V : Any, D : Comparable<D>> V.equalsVector(
|
|||||||
*/
|
*/
|
||||||
public fun Float64Vector2D.equalsVector(
|
public fun Float64Vector2D.equalsVector(
|
||||||
other: Float64Vector2D,
|
other: Float64Vector2D,
|
||||||
precision: Double = Euclidean3DSpace.defaultPrecision,
|
precision: Double = Float64Space2D.defaultPrecision,
|
||||||
): Boolean = equalsVector(Float64Space2D, other, precision)
|
): Boolean = equalsVector(Float64Space2D, other, precision)
|
||||||
|
|
||||||
/**
|
/**
|
||||||
@ -30,7 +35,7 @@ public fun Float64Vector2D.equalsVector(
|
|||||||
*/
|
*/
|
||||||
public fun Float64Vector3D.equalsVector(
|
public fun Float64Vector3D.equalsVector(
|
||||||
other: Float64Vector3D,
|
other: Float64Vector3D,
|
||||||
precision: Double = Euclidean3DSpace.defaultPrecision,
|
precision: Double = Float64Space3D.defaultPrecision,
|
||||||
): Boolean = equalsVector(Float64Space3D, other, precision)
|
): Boolean = equalsVector(Float64Space3D, other, precision)
|
||||||
|
|
||||||
/**
|
/**
|
||||||
|
@ -6,6 +6,7 @@
|
|||||||
package space.kscience.kmath.geometry
|
package space.kscience.kmath.geometry
|
||||||
|
|
||||||
import space.kscience.kmath.complex.Quaternion
|
import space.kscience.kmath.complex.Quaternion
|
||||||
|
import space.kscience.kmath.complex.QuaternionAlgebra
|
||||||
import space.kscience.kmath.complex.normalized
|
import space.kscience.kmath.complex.normalized
|
||||||
import space.kscience.kmath.geometry.euclidean3d.*
|
import space.kscience.kmath.geometry.euclidean3d.*
|
||||||
import space.kscience.kmath.structures.Float64Buffer
|
import space.kscience.kmath.structures.Float64Buffer
|
||||||
@ -26,12 +27,31 @@ class RotationTest {
|
|||||||
}
|
}
|
||||||
|
|
||||||
@Test
|
@Test
|
||||||
fun matrixConversion() {
|
fun matrixConversion() = with(QuaternionAlgebra){
|
||||||
|
|
||||||
val q = Quaternion(1.0, 2.0, -3.0, 4.0).normalized()
|
val q = Quaternion(1.0, 2.0, -3.0, 4.0).normalized()
|
||||||
|
|
||||||
val matrix = q.toRotationMatrix()
|
val matrix = q.toRotationMatrix()
|
||||||
|
|
||||||
|
for (ro in listOf(
|
||||||
|
RotationOrder.XYZ,
|
||||||
|
RotationOrder.YXZ,
|
||||||
|
RotationOrder.ZXY,
|
||||||
|
RotationOrder.ZYX,
|
||||||
|
RotationOrder.YZX,
|
||||||
|
RotationOrder.XZY
|
||||||
|
)) {
|
||||||
|
val angles = AngleVector.fromRotationMatrix(matrix, ro)
|
||||||
|
|
||||||
|
val reconstructed = Quaternion.fromEuler(angles, ro)
|
||||||
|
|
||||||
|
if( reconstructed.w>0) {
|
||||||
|
assertBufferEquals(q, reconstructed)
|
||||||
|
} else{
|
||||||
|
assertBufferEquals(q, -reconstructed)
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
assertBufferEquals(q, Quaternion.fromRotationMatrix(matrix))
|
assertBufferEquals(q, Quaternion.fromRotationMatrix(matrix))
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -50,4 +70,5 @@ class RotationTest {
|
|||||||
val q1 = Quaternion.fromEuler(0.1.radians, 0.2.radians, 0.3.radians, RotationOrder.XYZ)
|
val q1 = Quaternion.fromEuler(0.1.radians, 0.2.radians, 0.3.radians, RotationOrder.XYZ)
|
||||||
assertBufferEquals(Float64Buffer(0.9818562, 0.0640713, 0.0911575, 0.1534393), q1)
|
assertBufferEquals(Float64Buffer(0.9818562, 0.0640713, 0.0911575, 0.1534393), q1)
|
||||||
}
|
}
|
||||||
|
|
||||||
}
|
}
|
Loading…
Reference in New Issue
Block a user