lecture 05: exponential coordinates & homogeneous
TRANSCRIPT
Lecture 05:Exponential Coordinates &
Homogeneous TransformationsProf. Katie Driggs-Campbell
Co-Teaching with Prof. Belabbas
September 7, 2021
The First Friendly Robot
β’ In 1923, Prof. Makoto Nishimura saw R.U.R., who became deeply troubled with the idea that humans would struggle against artificial life β’ He was likely influenced by the increasingly popular automata,
mechanical figures that would exhibit autonomous behaviors
β’ He feared that humanity would be βdestroyed by the pinnacle of its creationβ
β’ Nishimura decided to create a different kind of artificial human that would celebrate nature, called Gakutensokuβ’ Nishimura refused to call his creation a robot. He named it
Gakutensoku, meaning βlearning from the rules of natureβ
β’ Disappeared after touring through Japan, China, and Korea
β’ Gakutensoku had a longstanding impact on Japanese robotics: many robot makers believe that machines are not in opposition to nature but rather part of it
https://spectrum.ieee.org/robotics/humanoids/the-short-strange-life-of-the-first-friendly-robot
This enlightened deity would be revealed βseated on agilded pedestal with its eyes closed, seemingly lost inthought. In its left hand, it holds an electric light with acrystal-shaped bulb, which it slowly lifts into the air. β¦ Thelightbulb illuminates, the giant opens its eyes, as if having arealization. Seemingly pleased with its eureka moment, itsmiles. Soon, it shifts its attention to a blank piece of paperlying in front of it and begins to write, eagerly recording itsnewfound insights.β
Administrivia
β’ Homework 2 is due Friday 9/10 at 8pm
β’ Project Update 0 will be due 9/11 at midnight
Angular Velocities
Angular Velocities in Reference Frame
Some useful properties and relations
Given any π β β3 and π β ππ 3 , the following holds: π π π β€ = π π
Now recall: ππ = αΆπ π β€
If π is π π π, we have that ππ = π ππ β ππ = π β€ππ
ππ = π β€ππ = π β€ αΆπ π β€ π = π β€ αΆπ = π β1 αΆπ
This gives us: ππ = αΆπ π β€ and ππ = π β€ αΆπ
Introducing Exponential Coordinates
Recall the Matrix Exponential
If αΆπ₯ π‘ = π΄π₯ π‘ , π₯ 0 = π₯0, where π₯ β βπ, π΄ β βπΓπ, the solution to the ODE is:
π₯ π‘ = ππ΄π‘π₯0The matrix exponential / the exponential function can be expanded:
ππ΄π‘ = πΌ + π΄π‘ +1
2π΄2π‘2 +
1
3!π΄3π‘3 +β―
Properties:
1.π
ππ‘ππ΄π‘ = π΄ππ΄π‘ = ππ΄π‘π΄
2. If π΄ = ππ·πβ1, then ππ΄π‘ = πππ·π‘πβ1
3. If π΄π΅ = π΅π΄, then ππ΄ππ΅ = ππ΅ππ΄ = ππ΄+π΅
4. Invertible: ππ΄π‘ β1 = πβπ΄π‘
Exponential Coordinate Representationβ’ Instead representing orientation as a rotation matrix, we introduce a three-
parameter representation: exponential coordinates
β’ Recall: ΰ·π rotation axis, π angle of rotation
β’ Then ΰ·ππ β β3gives the exponential coordinate representation
β’ A new interpretation for a frame coincident with π :β’ A rotation for 1 second around ΰ·π at angular velocity π, then the resulting frame is π
β’ A rotation for π seconds around ΰ·π at angular velocity 1, then the resulting frame is π
Exponential Coordinates of Rotations
Matrix Logarithm
β’ Given rotation matrix π , we need to take the logarithm to find the exponential coordinates:
β’ If we expand Rodrigues formula:Rot ΰ·π, π = πΌ + sin π ΰ·π + 1 β cos π ΰ·π 2
Rot ΰ·π, π
=
ππ + ΰ·π12(1 β ππ) ΰ·π1 ΰ·π2(1 β ππ) β ΰ·π3π π ΰ·π1 ΰ·π3 1 β ππ + ΰ·π2π π
ΰ·π1 ΰ·π2(1 β ππ) + ΰ·π3π π ππ + ΰ·π22 1 β ππ ΰ·π2 ΰ·π3(1 β ππ) β ΰ·π1π π
ΰ·π1 ΰ·π3(1 β ππ) β ΰ·π2π π ΰ·π2 ΰ·π3(1 β ππ) + ΰ·π1π π ππ + ΰ·π32 1 β ππ
exp: ΰ·π π β π π(3) β π β ππ(3)
log: π β ππ(3) β ΰ·π π β π π(3)
Matrix Logarithm Method
β’ If we take the trace of the matrix, we can solve for π:β’ π‘π π β π11 + π22 + π33 = 1 + 2 cos π
β’ If we compute π β€ β π , we get:β’ π32 β π23 = 2ΰ·π1 sin π
β’ π13 β π31 = 2ΰ·π2 sin π ΰ·π =1
2 sin ππ β€ β π
β’ π21 β π12 = 2ΰ·π3 sin π
β’ But what if sin π = 0? (called singularities)β’ If π = 2ππ, we have rotated by 360 degrees
β’ If π = (2π + 1)π, then Rodrigues formula is π = πΌ + 2 ΰ·π 2, which gives:
ΰ·ππ = Β±πππ+1
2and 2ΰ·ππ ΰ·ππ = πππ for π β π
Homogeneous Transformations: SE(3)
The special Euclidean group πΊπ¬(π) is the set of 4 Γ 4 matrices of the form:
π = π π , π =π π0 1
β’ The inverse of π is πβ1 = π β€ βπ β€π0 1
β ππΈ(3)
β’ If π1 and π2 β ππΈ(3), then π1π2 β ππΈ(3)
β’ π satisfies: ππ₯ β ππ¦ = π₯ β π¦ and
ππ₯ β ππ§ β€ ππ¦ β ππ§ = π₯ β π§ β€ π¦ β π§
Representing a configuration with SE(3)
Each frame can represent a body frame in a multi-link mechanism
As before:
ππππππ = πππππππ£π = π£π
Displacing Frames (1)
Example Displacement (1)
Displacing Frames (2)
Example Displacement (2)
Mobile arm exampleβ’ Robot arm mounted on wheeled platform.
Camera fixed to ceiling.
β’ {b} is body frame, {c} end-effector frame, {e} frame of object, and {a} is fixed frame.
β’ We assume the camera position and orientation in {a} is given.
β’ From camera measurements, you can evaluate the position and orientation of the body and the object in the camera frame.
β’ Since we designed our robot and have joint-angle estimates, we can obtain the end-effector position and orientation in the body frame.
β’ To pick up the object, we need the object position and orientation in the frame of our end-effector.
Summary
β’ Introduced exponential coordinates that allow us to parameterize rotations by the rotation axis ΰ·π and the angle of rotation π
β’ Walked through the matrix logarithm method, which tells us how to recover the exponential coordinates from a rotation matrix
β’ Started discussing homogeneous transformations where we not only look at rotations, but also translations in a single operation