L2-1 LQR

CDS 110b: Lecture 2-1
Linear Quadratic Regulators
Richard M. Murray
11 January 2006
Goals:
• Derive the linear quadratic regulator and demonstrate its use
Reading:
• Friedland, Chapter 9 (different derivation, but same result)
• RMM course notes (available on web page)
• Lewis and Syrmos, Section 3.3
Homework #2
• Design LQR controllers for some representative systems
• Due Wed, 18 Jan by 5 pm, in box outside 109 Steele
11 Jan 06 R. M. Murray, Caltech 1

Review from last lecture

Trajectory
Controller Process Estimator
Generation
Trajectory Generation via Optimal Control:
Today: focus on special case of a linear quadratic regulator

Linear Quadratic Regulator (finite time)
Problem Statement
• Factor of 1/2 simplifies some math below; optimality is not affected
Solution: use the maximum principle
• This is still a two point boundary value problem hard to solve

• Note that solution is linear in x (because is linear in x, treated as an input)

Simplified Form of the Solution
Can simplify solution by guessing that = P(t) x(t)
From maximum
principle
Substitute
= P(t) x(t)
Solution exists if we can find P(t) satisfying
• This equation is called the Riccati ODE; matrix differential equation

• Can solve for P(t) backwards in time and then apply u(t) = -R-1 B P(t) x
• Solving x(t) forward in time gives optimal state (and input): x*(t), u*(t)
• Note that P(t) can be computed once (ahead of time) allows us to find the
optimal trajectory from different points just by re-integrating state equation
with optimal input

Finite Time LQR Summary
Trajectory
Generation X
Problem: find trajectory that minimizes
Solution: time-varying linear feedback
• Note: this is in feedback form can actually eliminate the controller (!)

Infinite Time LQR
Extend horizon to T = and eliminate terminal constraint:
Solution: same form, but can show P is constant

State feedback (constant gain)
Algebraic Riccati equation
Remarks
• In MATLAB, K = lqr(A, B, Q, R)
• Require R > 0 but Q 0 + must satisfy “observability” condition
• Alternative form: minimize “output” y = H x
• Require that (A, H) is observable. Intuition: if not, dynamics may not affect
cost ill-posed. We will study this in more detail when we cover observers

Applying LQR Control
Trajectory
Generation
Application #1: trajectory generation

• Solve for (xd, yd) that minimize quadratic cost over finite horizon (requires
linear process)
• Use local controller to regulate to desired trajectory
Application #2: trajectory tracking

• Solve LQR problem to stabilize the system to the origin feedback u = K x
• Can use this for local stabilization of any desired trajectory
• Missing: so far, have assumed we want to keep x small (versus x xd)

LQR for trajectory tracking
Goal: design local controller to track xd:

ud
Trajectory xd e v
Generation
x
Approach: regulate the error dynamics

• Let e = x - xd, v = u - ud and assume f(x, u) = f(x) + g(x) u (simplifies notation)
• Now linearize the dynamics around e = 0 and design controller v = K e

• Final control law will be u = K (x - xd) + ud
• Note: in general, linearization will depend on xd u = K(xd)x “gain scheduling”

Choosing LQR weights
Most common case: diagonal weights
• Weight each state/input according to how much it contributes to cost

• Eg: if error in x1 is 10x as bad as error in x2, then q1 = 10 q2
• OK to set some state weights to zero, but all input weights must be > 0
• Remember to take units into account: eg for ducted fan if position error is in
meters and pitch error is in radians, weights will have different “units”
Remarks
• LQR will always give a stabilizing controller, but no gauranteed margins
• LQR shifts design problem from loop shaping to weight choices
• Most practical design uses LQR as a first cut, and then tune based on
system performance

Example: Ducted Fan

Stabilization:
• Given an equilibrium
position (xd, yd) and
( x, y ) equilibrium thrust f2d,
maintain stable hover
f2
• Full state available
f1 for feedback
Tracking:
• Given a reference
Equations of motion trajectory (xr(t), yr(t)),
find a feasible trajec-
tory and a
controller u = (x, xd,
ud) such that x xd
LQR design: see lqr_dfan.m (available on course web page)

Variation: Integral Action

ud
Trajectory xd e v
Generation
x
Limitation in LQR control: perfect tracking requires perfect model

• Control law is u = K (x - xd) + ud. ud must be perfect to hold e = 0
• Alternative: use integral feedback to give zero steady state error
integral of (output) error
• Now design LQR controller for extended system (including integrator weight)
equilibrium value y = r 0 steady state error

Example: Cruise Control
Linearized around v0:
Step 1: augment linearized (error) dynamics with integrator
Step 2: choose LQR weights and compute LQR gains Note: linearized
about v0 but try
to maintain
speed r (near v0)
Step 3: implement controller
PI controller

Summary: LQR Control
Trajectory
Generation
Application #1: trajectory generation

• Solve for (xd, yd) that minimize quadratic
cost over finite horizon
• Use local controller to track trajectory
Application #2: trajectory tracking

• Solve LQR problem to stabilize the system
• Solve algebraic Riccati equation to get state gain
• Can augment to track trajectory; integral action

Announcements
Mailing list
• If you didn’t get e-mail about TA office hours, send email to murray@cds
Late homework policy

• No late homework with out prior permission
• Usually willing to give a few extra days the first time you ask
• Sickness, conferences and other unavoidable conflicts usually work
Lecture recordings
• Will be posting audio recordings of lectures (along with slides) on web site

L2-1 LQR

Uploaded by

Copyright:

Available Formats

L2-1 LQR

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

L2-1 LQR

Uploaded by

Copyright:

Available Formats

CDS 110b: Lecture 2-1

Linear Quadratic Regulators

11 Jan 06 R. M. Murray, Caltech 1

Trajectory Generation via Optimal Control:

Today: focus on special case of a linear quadratic regulator

11 Jan 06 R. M. Murray, Caltech 2

• Factor of 1/2 simplifies some math below; optimality is not affected

Solution: use the maximum principle

• This is still a two point boundary value problem hard to solve

11 Jan 06 R. M. Murray, Caltech 3

Solution exists if we can find P(t) satisfying

• This equation is called the Riccati ODE; matrix differential equation

11 Jan 06 R. M. Murray, Caltech 4

Problem: find trajectory that minimizes

Solution: time-varying linear feedback

11 Jan 06 R. M. Murray, Caltech 5

Solution: same form, but can show P is constant

11 Jan 06 R. M. Murray, Caltech 6

Application #1: trajectory generation

Application #2: trajectory tracking

11 Jan 06 R. M. Murray, Caltech 7

Approach: regulate the error dynamics

• Now linearize the dynamics around e = 0 and design controller v = K e

11 Jan 06 R. M. Murray, Caltech 8

• Weight each state/input according to how much it contributes to cost

11 Jan 06 R. M. Murray, Caltech 9

LQR design: see lqr_dfan.m (available on course web page)

11 Jan 06 R. M. Murray, Caltech 10

Limitation in LQR control: perfect tracking requires perfect model

integral of (output) error

equilibrium value y = r 0 steady state error

11 Jan 06 R. M. Murray, Caltech 11

Step 1: augment linearized (error) dynamics with integrator

11 Jan 06 R. M. Murray, Caltech 12

Application #1: trajectory generation

Application #2: trajectory tracking

11 Jan 06 R. M. Murray, Caltech 13

Late homework policy

11 Jan 06 R. M. Murray, Caltech 14

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.