## 3.2.3. Exponential Coordinates of Rotation (Part 2 of 2)

#### 3.2.3. Exponential Coordinates of Rotation (Part 2 of 2)

This video describes how the solution of a vector linear differential equation calculates the rotation achieved after rotating a given time at a constant angular velocity. The matrix exponential maps the so(3) matrix representation of the 3-vector of exponential coordinates of rotation to a rotation matrix in SO(3), and the matrix logarithm maps a rotation matrix in SO(3) to the so(3) matrix representation of the 3-vector of exponential coordinates.

In this video we'll see how the matrix exponential can be applied to integrate the angular velocity of a rotating rigid body.

Let's start with this frame of coordinate axes, which will rotate about a unit angular velocity axis. To understand the motion of the coordinate axes, it suffices to consider just one of the coordinate axes, since the same reasoning applies to any axis. Let's call this remaining vector p.

As the vector p rotates about the rotation axis, it traces out a circle. The purpose of this video is to determine the final location of the vector if it rotates an angle theta about the rotation axis. We will do this by integrating the differential equation of motion describing the motion of p.

Here is a picture of our initial vector, p at time 0, and the unit rotation axis omega-hat.

As p begins to rotate, it traces out a circle around the rotation axis. The 3-vector linear velocity is tangent to the circle at any time, and is given by omega-hat cross p. After rotating an angle theta, the vector ends up at p at time theta.

At any instant of time, the time derivative of p is given by p-dot = omega-hat cross p. We can write this as a differential equation p-dot of t equals omega-hat cross p of t. The angular velocity is constant. Using our 3 by 3 skew-symmetric matrix notation, this becomes p-dot of t equals bracket omega-hat times p of t. This is a vector differential equation, whose solution, as we saw in the last video, is calculated using the matrix exponential. In general, a matrix exponential can be calculated using a series expansion, but when the matrix is 3 by 3 and skew symmetric, the series expansion has a simple closed form: the 3 by 3 identity matrix plus sin of theta times bracket omega-hat plus 1 minus cosine of theta times bracket omega-hat squared. In other words, the matrix exponential takes the skew-symmetric representation of the exponential coordinates omega-hat theta and calculates the corresponding rotation matrix. This equation is often called Rodrigues' formula.

Essentially, exponentiation integrates the angular velocity omega-hat for time theta seconds, going from the identity matrix to the final rotation matrix R. We can also define the inverse of the matrix exponential, the matrix logarithm, which takes a rotation matrix R and returns the skew-symmetric matrix representation of the exponential coordinates that achieve it, starting from the identity orientation. Just as the matrix exponential is like integration, the matrix log is like differentiation: it returns the unit angular velocity and the integration time that achieves the rotation matrix R. The matrix log is an algorithm that inverts Rodrigues' formula.

Later, when we're studying the kinematics of robots, the matrix exponential and log will become very useful. Basically, for a revolute joint, the unit angular velocity omega-hat represents the axis of rotation of the joint, and theta represents how far that joint has been rotated. Before we get to robot kinematics, though, we have to generalize the matrix exponential and log to cases where frames both rotate AND translate. In other words, general rigid-body motion. We'll start the process of generalizing from rotations to general rigid-body motions in the next video.