Translation, Rotation, and Transformation notes

Translations (Simple, Linear, Commutative) X Y Z 1 Dx Dy Dz Dy Dx

Rotations Differ from Translations Rotations are non-Euclidean like travelling on a globe vs. a grid Rotations are not commutative x-rotate, y-rotate is not equal y-rotate, x-rotate etc. Rotations are non-linear Rotations are not very intuitive, because when it comes to space, we think linearly. For example, if you get in a rocket ship and start moving to the left, you’ll keep going forever, never revisiting anyplace you’ve been, unless you change directions. This seems correct. However, if I continuously increase a rotation, I will keep coming back to my starting orientation an infinite number of times, because a rotation of zero degrees is equal to a rotation of 360 degrees. This is similar to moving on a sphere… and is a consequence of rotations being non-Euclidean. Secondly, if I give someone directions for traveling on a grid, and tell them their destination is 5 blocks to the left and 4 blocks up from here, it doesn’t matter whether they first go 5 blocks over and then 4 blocks up, or first go 4 blocks up and then five blocks over. But rotating first by 40 degrees around the X axis and then 50 degrees around the Y axis is very different from rotating 50 degrees around the Y axis and then 40 degrees around the X axis. And finally, rotations are non-linear functions of their parameters, whereas translations are linear functions. OK, so rotations are different from more intuitive translations in a number of ways. Why is that a problem?

Basic Rotation about Z-axis ú û ù ê ë é = 1 cos sin -sin j z R

Rotation Parameterization Represent rotation space in Euclidean R3 e.g. Euler angles Pros three parameters for three DOFs Cons singularities, potentially poor interpolation Parameterizations of rotations fall into two major groups - those that use Euclidean params, and those that use non-Euclidean. The first group represents a non-Euclidean 3 DOF rotation with 3 Euclidean parameters by “flattening” the rotation space. This sounds strange, but it’s very similar to what map-makers do when they attempt to represent the surface of the Earth on a grid by distorting the geometry as you get close to the poles. We’ll examine two different parameterizations that take this approach: Euler angles and the exponential map. The advantages to this approach are that we need only three parameters to represent three rotational DOFs, and our tools are happy because our parameterizations are Euclidean. Unfortunately, there are also problems, namely singularities and potential problems when we interpolate in these parameter spaces. We’ll examine this problems in detail when we talk about the individual parameterizations.

Euler angles (φ,θ,ψ) An Euler angle is a rotation about a single Cartesian axis Create multi-DOF rotations by concatenating Eulers R = Rψ Rθ Rφ 3 DOFs can be obtained by concatenating: An Euler angle is a rotation about a single Cartesian axis We can Create multi-DOF rotations by concatenating Eulers For example, we Can get a three DOF by concatenating X, Y, and Z Euler angles together. In doing so the The order in which the rotations are applied must remain fixed: always rotate first about X, then Y, then Z.

X-Convention Most commonly used The rotation given by Euler angles (φ,θ,ψ), where the first rotation is by an angle φ about the z-axis, the second is by an angle θ about the x-axis, and the third is by an angle ψ about the z-axis (again). R = Rψ Rθ Rφ ú û ù ê ë é = 1 cos sin -sin j R ú û ù ê ë é = q cos sin -sin 1 R ú û ù ê ë é = 1 cos sin -sin y R

Yaw-Pitch-Roll Convention ú û ù ê ë é = y cos sin -sin 1 R ú û ù ê ë é = 1 cos sin -sin j R ú û ù ê ë é = q cos -sin 1 sin R

Singularities More than one sets of parameters can create the same rotation matrix. Gimbal lock - two or more axes align, results in loss of rotational DOFs For Yaw-Pitch-Roll Convention As we mentioned previously, the Euler angle parameter space has singularities. The definition of a singularity is a continuous subspace of the entire parameter space, all of whose elements correspond to the same value (in this case, rotation). To continue our analogy to globes and maps, singularities occur at the north and south poles - each is but a single point on the globe, but when the map is unwrapped as we talked about before, the entire line at the top of the map corresponds to the north pole, and similarly for the south pole. Why is this bad? Because singularities induce gimbal lock - a state familiar to mechanical engineers in which two or more rotational axes lign up, resulting in a loss of a rotational DOF. What does that mean?! Let’s see...

Rotation Axis + Angle Euler’s Rotation Theorem: all rotations can be expressed as axis/angle

Rotation Matrix V = (1-Cos[q]) C = Cos[q] S = Sin[q] For given axis U(unit length) = {u1, u2, u3}T and rotation angle q u12 V + C u1u 2 V – u3 S u1u3 V + u 2S R = u 2u1 V + u3 S u 22 V + C u 2u3 V – u1 S u3u1 V – u2 S u3u 2 V + u1 S u32 V + C

Solution of Axis and Angle Sin[q] = ½{(R32-R23)2+(R13-R31)2+(R21-R12)2}(1/2) Cos[q] = (Trace[R]-1) / 2 q = Atan2(Sin[q], Cos[q]) -p < q < p y u1 = (R32-R23) / (2 Sin[q]) All Sine u2 = (R13-R31) / (2 Sin[q]) x Tan Cosine u3 = (R21-R12) / (2 Sin[q])

Transformation AP = ATB BP

Example: 1 5 -1 ATB = zA 5 xB yA yB xA zB

Cube of Sides 2 -1 2 -S[45] S[45] -C[45] 1 ATB = yB xB yA xA

Multiple Transformations AP = ATB BTC CTDDP Also DTA = (ATD)T

EOM’s L I q mg Newton SF=ma, SM=Ia SM=Ia - m g L Sin[q] = I a

EOM’s Lagrangian L = K – P (kinetic and potential energy) L q mg

EOM’s