tics - hichristensen.com€¦ · hip • If um, then • the e : •: tic que y • s end an tics...
Transcript of tics - hichristensen.com€¦ · hip • If um, then • the e : •: tic que y • s end an tics...
Manipulator Kinem
aticsForw
ard and Inverse Kinematics
2D Manipulator Forw
ardKinem
atics
•Forw
ard Kinematics
•G
iven T, find x
The vector of joint angles
The vector of end effector positions
Shorthand notation
2D Manipulator Inverse
Kinematics
•Inverse Kinem
atics = given x, find TThe joint angles
The end effector positions
Problem is m
ore difficult!
2D Manipulator Inverse Kinem
atics Example
•U
se the Law of Cosines:
•W
e also know that:
•D
o some algebra to get:
Again, note the notation short-hand
Inverse Kinematics Exam
ple Continued
•N
ow solve for c2:
•O
ne possible solution:
•Elbow
up vs elbow dow
n
•M
ay be impossible!
•M
ultiple solutions may exist!
•Fundam
ental problem in robotics!
Manipulator Kinem
aticsThe Jacobian
The Jacobian
•M
atrix analogue of the derivative of a scalar function
•A
llows us to relate end effector velocity to joint velocity
•G
iven
•The Jacobian, J, is defined as:
The Jacobian, a 2D 2-Link Manipulator
Example
•The forw
ard kinematics of a 2 link, revolute
joint manipulator are given as:
•W
hat is the relationship between the end
effector velocity, 𝒙, and the joint velocities,
𝜽?•
First step, take the time derivative of the
forward kinem
atics equations:
The Jacobian, a 2D 2-Link Manipulator
Example Continued
•Put into m
atrix form:
•W
e can further manipulate that to understand how
the relationship of the joints com
es into play:
Velocity of “elbow”
Velocity of “end effector”relative to “elbow
”
Singular Jacobians
•The Jacobian is singular w
hen its determinant is equal to 0.
•W
hen the Jacobian is non-singular the following relationship holds:
•Q
uestion---Intuitively, when is this not the case?
•H
int---Think of a configuration where changing the joints does not
change the end effector velocity in any arbitrary direction.
•W
orkspace boundaries
•D
emonstrate m
athematically?
•D
eterminant of the Jacobian = 0
Singular Jacobians Continued
•Find the determ
inant of the Jacobian:
•U
sing the following trig identities:
•Yields:
•Therefore,
•D
oes this make sense?
Singular Jacobians Continued
•The fact that
implies that all possible end effector
velocities are linear combinations of the follow
ing Jacobian matrix:
•M
atrix rank = # of linearly independent columns (or row
s)
•If the Jacobian is full rank, then the end effector can execute any arbitrary velocity
•Is this the case for our 2D
revolute joint planar manipulator?
Singular Jacobians Continued
•Look at the Jacobian again:
•W
e can see that it is made up of:
•W
here:
•The linear independence is a function of the joint angle
•If theta_2 = 0 or pi, the Jacobian loses rank
What is the physical m
anifestation of singularities?
•Configurations from
which certain instantaneous m
otions are unachievable•
A bounded end effector velocity (i.e. a velocity w
ith some finite value) m
ay correspond to an unbounded joint velocity (i.e. one that approaches infinity)
•if w
e want the end effector to attain a certain velocity, w
e need to input an infinite joint velocity
•N
ot a great idea!
•Bounded joint torques m
ay correspond to unbounded end effector torques and forces
•G
ood for static loading•
Not so good for other situations
•O
ften correspond to boundaries of the workspace
Summ
ary of Manipulator Kinem
atics Introduction
•Forw
ard kinematics is relatively sim
ple
•Inverse kinem
atics is relatively complicated and som
etimes
impossible
•A
Jacobian relates end effector velocity to joint velocity
•W
e typically want to com
pute the inverse of the Jacobian•
Typically we have a desired end effector velocity
•Then w
e need to compute the joint velocities to reach that end effector
velocity•
This requires us to compute the inverse of the Jacobian
•Som
etimes w
e can do this, sometim
es we cannot
•A m
anipulator’s singularities correspond to when the determ
inant of the Jacobian is zero.
Manipulator Kinem
aticsKinem
atic Optim
ization for Redundant Manipulators
Redundant Manipulators
Redundant Manipulators
•A
redundant manipulator is one w
here
•Jacobian represents the relationship betw
een end effector and joint velocities:
•For a given end effector velocity, to find joint velocities w
e can invert J
if it is square.
•W
ith a redundant manipulator, J is non-square so not invertible
•There m
ay exist different joint velocities, 𝜽, that yield the desired end
effector velocity, 𝒙•
Multiple solutions = opportunity to optim
ize!•
Minim
al energy•
Minim
al time
Redundant Manipulator –
Minim
al Energy
•M
inimize kinetic energy
•Kinetic energy is proportional to velocity squared
•Thus, w
e can minim
ize:
•W
here Wis a sym
metric
positive definite matrix
•W
eighting function•
Weighs im
portance of each joint•
Makes problem
more general
•Sym
metric = n
x nm
atrix such that
•A
is positive definite if zTAz
is positive for every non-zero colum
n vector zof n
real num
bers
•Exam
ple:
•If
•Then
Redundant Manipulator –
Minim
al Energy Continued
•M
inimize kinetic energy
•Subject to the kinem
atic constraint:
•To solve this problem
we need to em
ploy Lagrange Multipliers
Joseph-Louis Lagrange (born Giuseppe Lodovico
Lagrangia)
Lagrange Multipliers
•In an unconstrained
optimization problem
, we find the extrem
e points of som
e function:
•These are sim
ply where the partial derivatives are 0, or equivalently
where the gradient of f is 0:
where:
•In a constrained
optimization problem
, we add in another function:
Constrained Optim
ization Problem
•Tw
o methods to solve a constrained optim
ization problem•
Substitution•
Lagrange multipliers
Constrained Optim
ization Problem:
Substitution Example
•O
ptimize:
•Subject to:
•Solve for x
1in the constraint:
•Plug x
1into
to yield:
•W
e have transformed the tw
o dimensional constrained problem
into a one dim
ensional unconstrained problem
•D
ifferentiate as a function of x2 :
•Solve and substitute:
Constrained Optim
ization Problem:
Substitution Example Continued
•That w
as easy! Why do I have to learn about Lagrange m
ultipliers?
•The exam
ple was contrived
•G
enerally difficult•
Generally tim
e consuming
Constrained Optim
ization Problem: Lagrange
Multipliers Exam
ple
•The claim
: for an extremum
(minim
um or m
aximum
) of f(x) to exist on g(x), the gradient of f(x) m
ust be parallel to the gradient of g(x).
Find xand y
to maxim
ize f(x, y)subjectto a constraint (show
n in red) g(x, y) = c.
Images and text from
wikipedia
The red line shows the constraint g(x, y) =
c. The blue lines are contours of f(x, y). The point w
here the red line tangentially touches a blue contour is our solution. Since d
1> d
2 , the solution is a m
aximization of f(x, y)
Constrained Optim
ization Problem: Lagrange
Multipliers Exam
ple
•If the gradient of f(x) is parallel to the gradient of g(x), then the gradient of f(x) is a scalar function of the gradient of g(x):
•This m
eans the magnitude of the norm
al vectors do not need to be the sam
e, only the direction (the scalar, l, is term
ed the Lagrange m
ultiplier)•
This implies that:
•W
here:
Constrained Optim
ization Problem: Lagrange
Method Exam
ple
•O
ptimize:
•Subject to:
•W
e know that
Find thegradientConstrained O
ptimization Problem
: Lagrange M
ethod Example Continued
Plug into second equation, solve for
Plug x2 into third equation, solve for
Repeat for
Applying Lagrange Multipliers to the
Redundant Manipulator
•W
e have:
•W
here:
•Rem
ember that x
is m x 1 and q
is n x 1.
The function (energy) we w
ant to minim
ize, I substituted q for theta to represent a vector
Subject to the constraint(the m
anipulator kinematics)
Example: U
sing Lagrange Multipliers to solve
the Redundant Manipulator Problem
•Back to the problem
:
•W
is symm
etric positive definite, thus:
•Find the gradient:
•W
e want solve for the joint velocity in term
s of the end effector velocity, but w
e can't just invert Jbecause it is not square.
Example: U
sing Lagrange Multipliers to solve
the Redundant Manipulator Problem
Example: U
sing Lagrange Multipliers to solve
the Redundant Manipulator Problem
•Set the w
eighting function W to I
•W
here we can define
•A
s the generalized right-pseudo inverse of J.
Redundant Manipulators Sum
mary
•H
as more joint than endpoint degrees of freedom
•A
llows us to optim
ize the path of the robot
•U
se Lagrange multipliers to solve constrained optim
ization problem
Manipulator Kinem
aticsStatic Force Torque Relationships
Problem Statem
ent
•For a given force, F, acting on a m
anipulator’s end effector, what
torques, T, must be supplied at the joints to m
aintain static equilibrium
.
20th
Century Fox
Static Force/Torque Relationship
•Let:
•Let Gx and Gq represent infinitesim
al displacements in the task and
joint spaces (We w
ill call these virtual displacements) such that:
•The virtual w
ork of the system is given as:
•Rem
ember the Jacobian is:
Generalized force acting on the end effector
Static Force/Torque Relationship
•If the robot is in equilibrium
, then
•Com
bining the above equation with the Jacobian yields:
•Transposing both sides yields:
Static Force/Torque Relationship Summ
ary
•The joint torque required to hold a static force at a m
anipulator’s end effector is linearly proportional to the transpose of the Jacobian
Manipulator Kinem
aticsJoint Com
pliance
Problem Definition
•W
hat is the relationship between endpoint displacem
ent and an applied force at the endpoint m
anipulator for a given configuration of the m
anipulator?
•If I apply a force on the end effector, how
much w
ill the robot move?
•W
e know:
•A
ssume each joint ihas a stiffness, k
i :
Joint Compliance Derivation
Relationship between endpoint
displacement and endpoint force = C