Scribd
Upload
23 views

Introduction To Control - SSet15

This document provides an introduction to control systems for robotics. It discusses modeling robot dynamics using Lagrangian-Eulerian equations of motion and simplifying the model to first-order differential equations. Proportional-error (PE) control is introduced using positional error to generate torque commands. Issues with PE control like steady-state error and overshoot are addressed by adding proportional-derivative (PD) and then proportional-integral-derivative (PID) control terms. The state-space models of these control systems are derived. Adaptive and optimal control methods are also briefly discussed.

Uploaded by

Rishi Soni
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
23 views

Introduction To Control - SSet15

This document provides an introduction to control systems for robotics. It discusses modeling robot dynamics using Lagrangian-Eulerian equations of motion and simplifying the model to first-order differential equations. Proportional-error (PE) control is introduced using positional error to generate torque commands. Issues with PE control like steady-state error and overshoot are addressed by adding proportional-derivative (PD) and then proportional-integral-derivative (PID) control terms. The state-space models of these control systems are derived. Adaptive and optimal control methods are also briefly discussed.

Uploaded by

Rishi Soni
Copyright
© © All Rights Reserved
Available Formats
Download as PPT, PDF, TXT or read online on Scribd
You are on page 1/ 28

Introduction to Control:

How Its Done In Robotics


R. Lindeke, Ph. D.
ME 4135

Remembering the Motion Models:

As we found using the L-E approach, the


Required Joint Torque is:
gg

i Di (q) qi Ci (q, qi ) h(q) b(qi )


Dynamical
Manipulator
Inertial Tensor
a function of
position and
acceleration

Coupled joint
effects
(centrifugal and
coriolis) issues
due to multiple
moving joints

Gravitational
Effects
Frictional Effect
due to Joint/Link
movement

Lets simplify the model (a Bit)

This torque model is a 2nd order one (in position) lets look at it
as a velocity model rather than positional one then it becomes a
system of highly coupled 1st order differential equations

We will then isolate Acceleration terms (acceleration is the 1 st


derivative of velocity)

v q D (q ) i Ci (q, qi ) h(q ) b(qi )

gg

1
i

Considering Control:

Each Links torque is influenced by each other links


motion

We say that the links are highly coupled

Solution then suggests that control should come


from a simultaneous solution of these torques
We will model the solution as a State Space
design and try to balance the torque-in with
positional control-out the most common way it is
done!

But we could also use force control to solve the control


problem!

The State-Space Control Model:

State Space Approach

The State Space General Model feeds out positional


kinematics based on the torque (power) demand
input
Notice: 1/s is the Laplace transform of a unit step
(torque) impulse
As you remember, Laplace transforms convert linear
differential equation sets into algebraic equation sets

once solved we need to do inverse Laplace transforms to


return to torque/position space
In LaPlace models, S is a complex variable!

Ultimately how do we know How much Torque


to specify and if we are In Control

In robots with moving joints, we set


targets for motion
We sense motion at the joint level
(using kinesethic sensors)
We study differences between where
we are and where were going as a
Feedback (ie. servo) error
Control means that we try to minimize
error (make error go to zero) when we
move toward a new location

Setting up a Real Control

We will (start) by using positional error to drive our torque


devices

This simple model is called a PE (proportional error)


controller

PE Controller:

To a 1st approximation, = Km*I

Torque is proportional to motor current

And the Torque required is a function of Inertial


(Acceleration) and Friction (velocity) effects as
suggested by our L-E models

& Feq q&


m ; J eq q&
to the 1st approx:

& Feq q&


K m I m J eq q&

Setting up a Control Law

We will use the positional error (as drawn in the state model) to
develop our torque control
We say then for PE control:

k pe ( d a )

Here, kpe is a gain term that guarantees sufficient current will be


generated to develop appropriate torque based on observed
positional error

Using this Control Type:

It is a representation of the physical system of a


mass on a spring!
We say afer setting our target as a zero goal
that:
gg

k pe * a J F
a is a function
of the servo
feedback as a
function of time!

who's solution is:

a e

2J

t C e 12 t C e 12 t
2
1

Examining this solution:

The 1st term is a damping term for the motion

Ft

2J

The systems natural frequency is given by:

4 ke

2
J
J

Studying the solution:

If (F2/4kpe)> J we are over damped and the system will


never quite reach its target considering reasonable
time

If (F2/4kpe)= J we are critically damped giving the


system ideal behavior

If (F2/4kpe)< J we are under damped and the system will


over-shoot and oscillate about our target at the systems
natural frequency a dangerous situation in robotics!

These behaviors make sense


(physically!):
Under

High Friction Conditions (over


damped):

A system is hard to start but easy to stop

With High Moment of Inertia Conditions


(under damped):

A system is hard to start and it is hard to stop leading


to overshoot and possibly one that oscillates about the
target forever

Problems with PE Control:

First the so-called Steady-State


error the torque goes to zero when
the target is hit!

Secondly, we may be out of balance


the GAIN is not meeting our
Inertial vs. Friction balance leading
to overshoot or undershoot

Typically we will add a term to our model


to react to increasing speed so we
minimize overshoot

Dealing With Steady State Zero


Error

This is a gravitational issue: we


must add an L or gravitational
term back to our Torque control
model

Gravitational input is positionally


controlled:

L = -g (M1r1 + M2R)* Cos()


For a R Manipulator
with a payload on M2

Solving the Overshoot


Problem:

Lets expand our control law:

We should include a term that reacts to the


velocity of the link
But velocity is the derivative of the position
We will call this a proportional derivative
controller (PD Control)

k pe d a

d
kd
L
dt

State Space Model of PD:

Solving the PD Model:

Remembering:
gg

J eq q Feq q

pe
d
a

kd L J eq Feq

gg

k pe J eq Feq kd L
gg

Leads to a Solution of the


form:
Feq kd

t
2 J eq

C1e

.5 t

C2 e

where:

F k
eq
d

2
eq

4k pe

eq

.5 t

Effect of Derivative Term:

It is observed to be a form of
Active Friction
It tends to slow down the link as it
moves faster when high errors
(being far from goal) are observed
Thus it can be thought of as a
brake on the motion
It is a component that anticipates
changes and provides very fast
response to these changes

Taking Care of Trouble:

We add an integrator to the model

To damp out oscillations from over shoot


To control effects due to environmental
perturbations
To damp out Wild data gyrations typically due to
encoder errors

This leads to a model of control:


t

(t ) k pe ( d a t ) ki ( d a t )dt kd
t0

PID State Space Model:

Developing Optimal Control


PID

is most often found in the systems Arm


Joints
The components of Torque are functions of
POSE meaning the Jeq and Feq as well as L
factors change as one observes the robot
moving throughout the work envelope
We achieve control with kpe, kd and ki

if they are fixed values, we can expect critically


damped control at only a single (or very few)
position(s)

What is done:

We operate off critical PID on arm joints and PD


w/ gravitational compensation for remote and
wrist joints
Most controls use a form of adaptive control

Tables of gains applicable over certain geometries with


automatic changes as the manipulator moves about
the work envelope
We swap in and out the gain values such that we
minimize energy consumed by the drive:

min u 2 dt
given by :

u k pe kd

Another Idea:
Develop

a Performance Index (PI) that


judges controller stability

This

PI is an external measurement scheme


that using logic and comparisons between
desired and actual performance then adjusts
the model

State Model of Adjustable


Controller

Thus Ends our Introductory


Studies of Robotics & Controls

This is a rich and deep field of applied


Mechanical and Industrial Engineering

While we have deeply explored some topics,


others have only been scanned

I wish you well as you move forward in your


lifelong exploration of AUTOMATION and its
myriad of supporting technologies

I sincerely hope that you have all learned


something of this fascinating field and that
these lessons will prove to be valuable in
your careers!

You might also like